system

Abstract

We provide the system. [Solution] A means for receiving a PDF of a paper from a user, A means for extracting text, images, and mathematical formulas from the aforementioned PDF, A method using a natural language processing model to analyze the extracted text, A means of using an image analysis model to analyze the extracted images, One method is to use a mathematical analysis tool to analyze the extracted mathematical formulas, A means for integrating the aforementioned analysis results and generating program code necessary for reproduction experiments, Means for providing the user with the analysis results and the generated program code, A system that includes this.

Description

Technical Field

[0002]

[0001] The technology of the present disclosure relates to a system.

Background Art

[0002] Patent Document 1 discloses a method for controlling a persona chatbot, which is performed by at least one processor, including steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to an explanation of a character of the chatbot, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Conventional paper reading support systems have been mainly limited to text analysis, and it has been difficult to analyze images and mathematical formulas included in papers. As a result, it has required a great deal of effort and high-level expertise to grasp the overall picture of the content described in the paper and conduct reproducible experiments. Due to such constraints, it has been difficult for researchers to conduct reproducible experiments on papers, and as a result, the progress of research has tended to be delayed. In particular, in papers where no code is provided, it has been even more difficult to reproduce the experiments. The present invention aims to solve these problems and enable an accurate and comprehensive understanding of the overall picture of a paper and easy conduct of reproducible experiments.

Means for Solving the Problems

[0005] The present invention is a system comprising: means for receiving a PDF of a paper from a user; means for extracting text, images, and mathematical formulas from the PDF; means for using a natural language processing model to analyze the extracted text; means for using an image analysis model to analyze the extracted images; means for using a mathematical formula analysis tool to analyze the extracted mathematical formulas; means for integrating the analysis results and generating program code necessary for replication experiments; and means for providing the analysis results and the generated program code to the user. This allows researchers to easily conduct replication experiments by comprehensively analyzing the content of a paper and providing the information necessary for replication experiments, thereby accelerating the progress of research.

[0006] A "user" is an individual or organization that uploads a PDF of a research paper to the system and receives the analysis results.

[0007] A "paper" is a PDF document that describes the results of academic research.

[0008] "PDF" stands for Portable Document Format, and it is a digital document that includes text and images while maintaining the original document format.

[0009] "Text" refers to the written data contained in a research paper.

[0010] "Images" refer to the data of figures, tables, and graphics included in a research paper.

[0011] A "mathematical formula" is data that represents the mathematical expression described in a paper.

[0012] A "natural language processing model" is a model that uses artificial intelligence technology to analyze text data and understand or generate its content.

[0013] An "image analysis model" is a model that uses artificial intelligence technology to analyze image data and identify objects and features within it.

[0014] A "mathematical formula analysis tool" is a tool that uses technologies to analyze mathematical formula data and understand or generate its content.

[0015] "Analysis result" refers to data obtained by analyzing text, images, and mathematical formulas.

[0016] "Integration" refers to combining multiple analysis results to create one comprehensive result.

[0017] "Replication experiment" refers to conducting the same or similar experiments based on the content of a paper.

[0018] "Program code" is a set of instructions described in a form executable by a computer.

[0019] "Provision" refers to passing data or information from a system to a user.

Brief Description of Drawings

[0020] [Figure 1] It is a conceptual diagram showing an example of the configuration of a data processing system according to the first embodiment. [Figure 2] It is a conceptual diagram showing an example of the main functions of a data processing device and a smart device according to the first embodiment. [[ID=​​​​​​​​​​​​​​​It is a conceptual diagram showing an example of the main functions of a data processing device and a robot according to the fourth embodiment. [Figure 9] It shows an emotion map to which a plurality of emotions are mapped. [Figure 10] It shows an emotion map to which a plurality of emotions are mapped. [Figure 11] It is a sequence diagram showing the processing flow of the data processing system in Example 1. [Figure 12] It is a sequence diagram showing the processing flow of the data processing system in Application Example 1. [Figure 13] It is a sequence diagram showing the processing flow of the data processing system in Example 2 when an emotion engine is combined. [Figure 14] It is a sequence diagram showing the processing flow of the data processing system in Application Example 2 when an emotion engine is combined.

Embodiments for Carrying Out the Invention

[0021] Hereinafter, an example of an embodiment of a system according to the technology of the present disclosure will be described with reference to the accompanying drawings.

[0022] First, the terms used in the following description will be explained.

[0023] In the following embodiments, a processor with a reference number (hereinafter simply referred to as "processor") may be a single arithmetic device or a combination of a plurality of arithmetic devices. Also, the processor may be a single type of arithmetic device or a combination of a plurality of types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), an APU (Accelerated Processing Unit), and the like.

[0024] In the following embodiments, signed RAM (Random Access Memory) is a memory that temporarily stores information and is used as work memory by the processor.

[0025] In the following embodiments, the signed storage is one or more non-volatile storage devices that store various programs and various parameters. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), or magnetic tapes.

[0026] In the following embodiments, the signed communication interface (I / F) is an interface that includes a communication processor and an antenna, etc. The communication interface manages communication between multiple computers. Examples of communication standards applicable to the communication interface include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0027] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." That is, "A and / or B" means that it may be A alone, or B alone, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" applies when expressing three or more things linked by "and / or."

[0028] [First Embodiment]

[0029] Figure 1 shows an example of the configuration of the data processing system 10 according to the first embodiment.

[0030] As shown in Figure 1, the data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0031] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0032] The smart device 14 comprises a computer 36, a reception device 38, an output device 40, a camera 42, and a communication interface 44. The computer 36 comprises a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The reception device 38, output device 40, and camera 42 are also connected to the bus 52.

[0033] The reception device 38 is equipped with a touch panel 38A and a microphone 38B, etc., and receives user input. The touch panel 38A receives user input by detecting contact with an object (e.g., a pen or finger). The microphone 38B receives user input by detecting the user's voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0034] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form perceptible to the user 20 (e.g., audio and / or text). The display 40A displays visible information such as text and images according to instructions from the processor 46. The speaker 40B outputs audio according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0035] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various types of information between processor 46 and processor 28 via network 54.

[0036] Figure 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0037] As shown in Figure 2, in the data processing device 12, a specific processing is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" related to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0038] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0039] In the smart device 14, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The reception output program 60 is used in conjunction with a specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0040] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the smart device 14 as the "terminal".

[0041] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. This system operates on a server side and analyzes PDFs uploaded by users, providing the results.

[0042] Program processing flow

[0043] 1. Receiving the PDF

[0044] Users upload their research papers as PDFs to the server via a web interface. This serves as the entry point to the system.

[0045] 2. Data Extraction

[0046] The server saves the received PDF file to a specific directory. Next, a PDF parsing library is used to extract text, images, and mathematical formulas from the PDF.

[0047] The text is read page by page, combined, and saved as a single continuous text file. Images are saved in a specific folder, and mathematical formulas are converted into text format using OCR technology.

[0048] 3. Text Analysis

[0049] The server utilizes a natural language processing model to analyze the extracted text. This generates a summary and key points of the paper.

[0050] 4. Image Analysis

[0051] The server uses an image analysis model to analyze images within the paper. It retrieves data points from graphs and figures contained within the images, as well as their interpretations, in text format.

[0052] 5. Formula Analysis

[0053] The server uses a mathematical analysis tool to analyze the extracted mathematical formulas. The analysis results are saved as text data, including interpretations of the formulas' meanings and applications.

[0054] 6. Support for Integration and Reproducibility Experiments

[0055] The server integrates the analysis results and generates code for replication experiments. This allows users to obtain specific procedures and program code to reproduce the experiment based on the content of the paper.

[0056] 7. Providing the results

[0057] The server returns the generated analysis results and program code to the user. The user can then use this to gain a deeper understanding of the paper's content and smoothly carry out replication experiments.

[0058] Specific example

[0059] As a concrete example, we can consider the following scenario.

[0060] 1. Receiving the PDF

[0061] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[0062] 2. Data Extraction

[0063] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[0064] 3. Text Analysis

[0065] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary such as "This paper proposes a new gradient descent method that converges faster than conventional methods" is generated.

[0066] 4. Image Analysis

[0067] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[0068] 5. Formula Analysis

[0069] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0070] 6. Support for Integration and Reproducibility Experiments

[0071] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[0072] 7. Providing the results

[0073] The server provides the user with a generated summary, image analysis results, mathematical formula interpretations, and code for replication experiments. The user can then use this information to reproduce the paper's experiments in their own environment.

[0074] As described above, the system of the present invention enables users to deeply understand the content of research papers and efficiently conduct reproduction experiments.

[0075] The following describes the processing flow.

[0076] Step 1:

[0077] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0078] Step 2:

[0079] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[0080] Step 3:

[0081] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[0082] Step 4:

[0083] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. The analysis results are saved in text format and used for subsequent processing.

[0084] Step 5:

[0085] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts the constituent elements and features of each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[0086] Step 6:

[0087] The server extracts mathematical formulas and analyzes them using a mathematical analysis tool (e.g., Mathpix). The formula images are converted into text format, and the meaning and application examples of the formulas are described in natural language. This allows the content of the formulas to be saved as text data.

[0088] Step 7:

[0089] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[0090] Step 8:

[0091] The server provides the user with the generated analysis results and program code. The user downloads this from the web interface and uses it as a reference to conduct reproduction experiments.

[0092] (Example 1)

[0093] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the smart device 14 will be referred to as the "terminal."

[0094] Conventional technologies required advanced expertise and considerable time to comprehensively analyze the content of papers and documents and facilitate reproduction experiments. Furthermore, the process of integrating results using multiple analysis methods and tools was complex and burdensome for users. Therefore, the present invention aims to provide a system that automates document analysis and preparation for reproduction experiments, enabling efficient and accurate execution.

[0095] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0096] In this invention, the server includes means for receiving a document file from a user, means for extracting character data, figures, and formulas from the file, means for using a natural language processing model to analyze the extracted character data, means for using an image analysis model to analyze the extracted figures, means for using an analysis tool to analyze the extracted formulas, means for integrating the analysis results and generating a program necessary for reproduction verification, and means for providing the user with the analysis results and the generated program. This makes it possible to comprehensively analyze the contents of a document and to perform reproduction experiments efficiently and accurately.

[0097] A "user" refers to an entity that uploads document files to the server and receives the analysis results or programs.

[0098] "Document" refers to electronic files containing text information such as papers and reports, including formats such as PDF.

[0099] A "file" refers to digital data in which the contents of a document are stored.

[0100] "Character data" refers to the text information contained within a document.

[0101] "Graphics" refers to visual information such as images, graphs, and charts included within a document.

[0102] "Formula" refers to mathematical formulas or mathematical expressions included in a document.

[0103] "Extraction" refers to the process of taking specific data (text data, graphics, formulas) out of a file.

[0104] A "natural language processing model" refers to a machine learning model or algorithm used to analyze extracted text data and extract important information.

[0105] An "image analysis model" refers to a machine learning model or algorithm that analyzes extracted graphic data and converts visual information into text format.

[0106] An "analysis tool" refers to software or algorithms used to analyze extracted formulas and understand their meaning.

[0107] "Integration" refers to the process of combining analysis results into a single format and ensuring consistency.

[0108] "Reproduction verification" refers to the process of reproducing experiments or simulations based on the content of a document.

[0109] A "program" refers to a set of code and procedures necessary for reproducing and verifying a problem.

[0110] "Results" refers to the information obtained from the analysis of the document.

[0111] This invention is a document analysis and reproduction experiment support system, primarily configured to operate on the server side. The specific implementation methods for each step are described below.

[0112] Receiving document files

[0113] Users upload document files to the server using a web interface. These document files include PDFs and other formats containing text information such as papers and reports. The server saves the received document files to a specific directory (e.g., / tmp / pdf_uploads / ).

[0114] Data extraction

[0115] The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text data, shapes, and formulas from the file. The specific steps are as follows:

[0116] Text data: Extracts text from each page of a document and saves it as a series of text data.

[0117] Shapes: Extracts images, graphs, and charts from a document and saves them, for example, to a folder such as / tmp / pdf_uploads / images / .

[0118] Formula: Use an OCR library (e.g., Mathpix) to convert mathematical formulas into text format and save them.

[0119] Data Analysis

[0120] Analysis of character data

[0121] The server uses natural language processing models (e.g., BERT, GPT) to analyze text data and extract summaries and key points from papers. This allows the server to generate summaries that enable users to quickly understand the content of documents.

[0122] Geometric analysis

[0123] The server analyzes the shapes using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format. The analysis results include, for example, interpretations of the model's architecture diagram and graphs of experimental results.

[0124] Analysis of the formula

[0125] The server uses a mathematical analysis tool (e.g., SymPy) to analyze mathematical formulas and generate interpretations of their meaning and applications. The analysis results are saved as text data explaining the meaning of the formulas.

[0126] Support for integration and replication experiments

[0127] The server integrates these analysis results and generates a program for replication experiments. This program is provided as Python code that includes model definition, data preprocessing, and training procedures. Users can use this program to replicate the experiment in their own environment.

[0128] Providing results

[0129] The server returns the generated analysis results and program code to the user. The analysis results include a summary, image analysis results, interpretation of mathematical formulas, and code for reproduction experiments, enabling the user to gain a deeper understanding of the document's content and efficiently conduct reproduction experiments.

[0130] Specific example

[0131] As a concrete example, consider a case where a user uploads a paper on a new learning method for AI models. The server can perform the analysis through the following steps:

[0132] The server extracts 40 pages of text, 5 images, and 3 mathematical formulas, and analyzes each of them.

[0133] The server uses a natural language processing model to generate a summary that reads, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[0134] The server uses an image analysis model to analyze the image, concluding that "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer."

[0135] The server uses a mathematical analysis tool to interpret Equation 1 as "showing the loss function of the proposed method, which minimizes the squared error."

[0136] The server generates Python code for reproduction experiments and provides it to the user.

[0137] Examples of prompt statements

[0138] "Analyze a paper on a new AI model learning method, extract key points, interpret image content and mathematical formulas, and generate code for replication experiments."

[0139] Please analyze the following PDF paper and provide the summary, data points for the graphs, meaning of the formulas, and experimental replication procedures. PDF link: [link]

[0140] As described above, the document analysis system of the present invention can provide an effective means for users to deeply understand the content of documents and efficiently conduct reproduction experiments.

[0141] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0142] Step 1:

[0143] PDF reception

[0144] Input: The PDF file to be uploaded by the user.

[0145] Specific operation: On the web interface, the user clicks the "Select File" button, selects the desired PDF file from their local disk, and clicks the upload button.

[0146] Data processing: The server saves the received PDF files to a specific directory (e.g., / tmp / pdf_uploads / ).

[0147] Output: PDF file saved in / tmp / pdf_uploads / .

[0148] Step 2:

[0149] Data extraction

[0150] Input: Saved PDF file.

[0151] Specific operation: The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text, images, and mathematical formulas from PDF files.

[0152] Text extraction: Reads text from each page and saves it as a series of text files (e.g., / tmp / pdf_uploads / extracted_text.txt).

[0153] Image extraction: Identifies images within the PDF and saves them to a specific folder (e.g., / tmp / pdf_uploads / images / ).

[0154] Formula extraction: Use an OCR library (e.g., Mathpix) to convert formulas in the PDF into text format and save it (e.g., / tmp / pdf_uploads / extracted_equations.txt).

[0155] Data processing: Extraction of text, images, and mathematical formulas using a PDF analysis library.

[0156] Output: Text data ( / tmp / pdf_uploads / extracted_text.txt), image data ( / tmp / pdf_uploads / images / ), mathematical formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[0157] Step 3:

[0158] Text analysis

[0159] Input: Extracted text data ( / tmp / pdf_uploads / extracted_text.txt).

[0160] Specific operation: The server uses a natural language processing model (e.g., BERT, GPT) to analyze the text data and extract summaries and key points.

[0161] Section Identification: Analyzes text data to identify sections of a paper (e.g., Introduction, Methods, Results, Discussion, etc.).

[0162] Summary generation: Generate summaries for each section and extract the core elements of the new learning method.

[0163] Data processing: Text analysis and summarization using natural language processing models.

[0164] Output: Summarized text data.

[0165] Step 4:

[0166] Image analysis

[0167] Input: Extracted image data ( / tmp / pdf_uploads / images / ).

[0168] Specific operation: The server analyzes images using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format.

[0169] Graph and chart analysis: Extracting data points and interpretations from images (e.g., "Figure 1 shows a 3-layer convolutional neural network, with ReLU used in each layer").

[0170] Data processing: Analysis of image data using image analysis models.

[0171] Output: Text data of the image analysis results.

[0172] Step 5:

[0173] Formula analysis

[0174] Input: Extracted formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[0175] Specific operation: The server uses a mathematical expression analysis tool (e.g., SymPy) to analyze the formula and generate an interpretation of its meaning and application.

[0176] Mathematical analysis: Analyze the mathematical formula and interpret it as "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0177] Data calculation: Analysis of mathematical formulas using mathematical analysis tools.

[0178] Output: Text data of the mathematical analysis results.

[0179] Step 6:

[0180] Support for integration and replication experiments

[0181] Input: Text analysis results, image analysis results, mathematical formula analysis results.

[0182] Specific operation: The server integrates these analysis results and generates a program for reproduction experiments.

[0183] Integration of analysis results: Organize the analysis results and compile them into a single document (e.g., Markdown format).

[0184] Program generation: Generates Python code for replication experiments, describing the model definition, data preprocessing, and training procedures.

[0185] Data processing: Integration of analysis results and generation of program code.

[0186] Output: Integrated analysis results document and Python code.

[0187] Step 7:

[0188] Providing results

[0189] Input: Integrated analysis results document and Python code.

[0190] Specific operation: The server compresses the generated analysis results and program code into a zip file and provides a link for the user to download it.

[0191] Output: A download link accessible to the user.

[0192] The above describes the specific operation and input / output of each processing step in this system.

[0193] (Application Example 1)

[0194] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server," and the smart device 14 will be referred to as the "terminal."

[0195] When new production technologies or methods are proposed within a factory, actually reproducing and verifying them requires time and effort. Furthermore, understanding technical documents such as research papers requires specialized knowledge, and appropriately interpreting their content and reflecting it in experiments is difficult. As a result, the introduction of new technologies is often delayed, hindering improvements in production efficiency and quality.

[0196] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[0197] In this invention, the server includes means for receiving a PDF of a technical document from a user; means for extracting information, images, and mathematical representations from the PDF; means for using a natural language processing model to analyze the extracted information; means for using an image processing model to analyze the extracted images; means for using a mathematical expression analysis tool to analyze the extracted mathematical representations; means for integrating the analysis results and generating program code and automated experimental procedures necessary for reproduction experiments; and means for providing the analysis results and generated program code to the user. This makes it possible to quickly analyze new technologies described in technical documents within a factory and to automatically perform reproduction experiments.

[0198] A "user" is an individual or organization that operates the system to analyze technical documents or conduct reproduction experiments.

[0199] "Technical documents" refer to documents containing content related to science, technology, or engineering, such as research papers and technical reports.

[0200] "PDF" is an abbreviation for Portable Document Format, a file format for electronically displaying the contents of a document.

[0201] "Information" refers collectively to the text, images, and mathematical formulas included in technical documents.

[0202] "Images" refer to visual data such as diagrams, graphs, and photographs included within technical documents.

[0203] "Mathematical expression" refers to mathematical formulas and statistical descriptions included in technical documents.

[0204] A "natural language processing model" is a computational model used to analyze text data and understand its meaning and structure.

[0205] An "image processing model" is a computational model used to analyze image data and extract its content and features.

[0206] A "mathematical formula analysis tool" is software used to analyze mathematical formulas and understand their meaning and structure.

[0207] "Analysis results" refers to the collective data obtained by natural language processing models, image processing models, and mathematical analysis tools.

[0208] "Program code" refers to a computer program that describes the procedures and processes necessary to carry out a reproduction experiment.

[0209] An "automated experimental procedure" is a specific set of steps generated based on analysis results, which are used to perform a reproducible experiment using automated equipment.

[0210] A "system" is a general term for a complex set of devices or software that automates everything from the analysis of technical documents to reproduction experiments.

[0211] This invention provides a system that analyzes PDF technical documents and comprehensively analyzes text, images, and mathematical representations to facilitate users in performing replication experiments. This system includes a paper decoding system, a server, a cloud server, and a factory robot.

[0212] System Overview

[0213] Program processing

[0214] The server first receives technical document PDFs from users. The received PDFs are saved in a specific directory. Next, a PDF parsing library (e.g., PyPDF2) is used to extract text, images, and mathematical representations from the PDFs. The extracted text is analyzed using a natural language processing model (e.g., spacy) to generate summaries and key points of the papers. Images are analyzed using an image processing model (e.g., pytesseract) to convert the included data and graphs into text format. Mathematical representations are analyzed using a mathematical expression analysis tool (e.g., regular expressions) to explain their meaning.

[0215] The analysis results are integrated to generate the program code and automated experimental procedures necessary for reproduction experiments. This generated code is provided to the user. Furthermore, the system provides procedures and tools for automatically performing reproduction experiments based on the analysis results.

[0216] Hardware and software used

[0217] Hardware:

[0218] Factory robots (e.g., KUKA robots, ABB robots)

[0219] Cloud servers (AWS, Google Cloud, etc.)

[0220] software:

[0221] PDF analysis: PyPDF2

[0222] Natural Language Processing: Spacy

[0223] Image analysis: pytesseract

[0224] Formula extraction: Regular expressions (re)

[0225] Specific example

[0226] As a concrete example, consider a scenario in which a technical document on a new, highly efficient production technology is introduced to a factory. First, the factory manager uploads the PDF of the technical document to a factory robot using a smartphone. Analysis is performed on a cloud server, and the following results are generated.

[0227] Summary: "This technical document proposes a new technology that achieves a 20% improvement in efficiency compared to conventional technology."

[0228] Image analysis: "Figure 1 shows a flowchart of the new technology, and the main steps are as follows..."

[0229] Mathematical analysis: "Equation 1 shows the optimization parameters for improving efficiency."

[0230] Based on these analysis results, detailed program code necessary for reproduction experiments is generated. This generated code is provided to the factory robot, which automatically executes the reproduction experiment. This process allows for rapid verification of the effectiveness of new technologies and supports their implementation within the factory.

[0231] Example of a prompt

[0232] Examples of specific prompt messages are as follows:

[0233] "Please write a program for industrial robots to analyze papers on new production technologies and automate the replication experiments."

[0234] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0235] Step 1:

[0236] PDF reception

[0237] The user uploads a PDF file of a technical document. This PDF is received by the server and saved to a specific directory.

[0238] Input: PDF file of a technical document.

[0239] Output: PDF file saved in the server directory.

[0240] Specific operation: When a user selects a PDF from the web interface and presses the upload button, the server saves the file to the specified directory.

[0241] Step 2:

[0242] Data extraction

[0243] The server uses the PDF parsing library (PyPDF2) to extract text, images, and mathematical representations from PDF files.

[0244] Input: Saved PDF file.

[0245] Output: Extracted text files, a list of image files, and a list of mathematical representations.

[0246] Specific operation: The server opens the PDF file, extracts text from each page, saves the image files, and adds the mathematical representations to a list.

[0247] Step 3:

[0248] Text analysis

[0249] The server uses a natural language processing model (SPacy) to analyze the extracted text.

[0250] Input: Extracted text file.

[0251] Output: Text summary and key points.

[0252] Specific operation: The server inputs a text file into a natural language processing model and extracts a summary and important sections.

[0253] Step 4:

[0254] Image analysis

[0255] The server uses an image processing model (pytesseract) to analyze the extracted images.

[0256] Input: List of extracted image files.

[0257] Output: Text data extracted from the image.

[0258] Specific operation: The server sequentially inputs each image file into the image processing model and extracts the text data.

[0259] Step 5:

[0260] Formula analysis

[0261] The server uses a mathematical expression analysis tool (regular expressions) to analyze the extracted mathematical representation.

[0262] Input: A list of mathematical expressions.

[0263] Output: Interpretation and explanation of the meaning of the formula.

[0264] Specific operation: The server inputs the list of mathematical formulas into a regular expression tool and generates text explaining the meaning of each formula.

[0265] Step 6:

[0266] Support for integration and replication experiments

[0267] The server integrates the analysis results and generates the program code and automated experimental procedures necessary for reproduction experiments.

[0268] Input: Text summaries, text data extracted from images, and interpretation of mathematical formulas.

[0269] Output: Integrated analysis results and program code for replication experiments.

[0270] Specific operation: The server integrates all data and generates Python code for reproduction experiments based on the analysis results.

[0271] Step 7:

[0272] Providing results

[0273] The server provides the user with the generated analysis results and program code.

[0274] Input: Integrated analysis results and program code for replication experiments.

[0275] Output: Analysis results and program code provided to the user.

[0276] Specific operation: The server formats the analysis results and code and provides them to the user as a download link. It also sends the analysis results via email or other methods as needed.

[0277] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0278] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system operates on a server side, analyzing PDFs uploaded by users and providing the results.

[0279] Program processing flow

[0280] 1. Receiving the PDF

[0281] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0282] 2. Data Extraction

[0283] The server receives a PDF file and temporarily saves it in storage. For example, save it as / tmp / uploaded_paper.pdf and pass the file path to the following process.

[0284] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Parse each page, concatenate the text into one continuous text, save the images in a separate directory, and save the mathematical formulas as images.

[0285] 3. Text Analysis

[0286] The server utilizes a natural language processing model (e.g., GPT-4) to analyze the extracted text. This generates an abstract and the main points of the paper. The analysis results are saved in text format and used for subsequent processing.

[0287] 4. Image Analysis

[0288] The server uses an image analysis model (e.g., YOLO, CNN) for analysis. Extract the components and features of the charts from each image and save the information obtained in text format. For example, extract the parameters of each layer and activation functions, etc.

[0289] 5. Mathematical Formula Analysis

[0290] The server uses a mathematical formula analysis tool (e.g., Mathpix) for analysis. Convert the mathematical formula image into text format and describe the meaning and application examples of the mathematical formula in natural language. Thereby, save the content of the mathematical formula as text data.

[0291] 6. Sentiment Recognition

[0292] The server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed through user input, interaction, and video analysis from the webcam. For example, it can determine in real time whether the user is showing anxiety or interest.

[0293] 7. Support for Integration and Reproducibility Experiments

[0294] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[0295] Furthermore, the method of providing analysis results will be adjusted based on the user's emotions. For example, if a user is feeling anxious, more detailed explanations or step-by-step guides will be provided.

[0296] 8. Providing the results

[0297] The server provides the user with the generated analysis results and program code. The order and method of presenting the results are adjusted according to the user's mood. For example, if the user appears interested, important results can be presented first to maintain their attention.

[0298] Users download this from the web interface and use it as a reference for conducting reproduction experiments. The reproduction experiment process proceeds based on the provided code and explanations.

[0299] Specific example

[0300] As a concrete example, we can consider the following scenario.

[0301] 1. Receiving the PDF

[0302] The user uploads the "Paper on a New Learning Method for AI Models" to the server. This PDF describes in detail the new method of AI research.

[0303] 2. Data Extraction

[0304] The server extracts from the PDF 40 pages of text, 5 images, and 3 complex mathematical formulas. Each image includes graphs and architecture diagrams of models.

[0305] 3. Text Analysis

[0306] The server uses a natural language processing model to summarize the main sections of the paper and extract the core part of the new learning method. A summary such as "This paper proposes a new gradient descent method, which converges faster than the conventional method" is generated.

[0307] 4. Image Analysis

[0308] The server extracts in text form the parameters and activation functions of each layer from the images. Information such as "Figure 1 shows a three-layer convolutional neural network, and ReLU is used in each layer" is obtained.

[0309] 5. Mathematical Formula Analysis

[0310] The server analyzes the mathematical formulas and generates text explaining the meaning of the formulas. It is interpreted as "Formula 1 shows the loss function of the proposed method, which minimizes the squared error".

[0311] 6. Emotion Recognition

[0312] The server recognizes the user's emotion and determines whether the user feels uneasy about the analysis results. For example, if the user has asked many questions about past analysis results, it is determined that the user feels uneasy.

[0313] 7. Support for Integration and Reproduction Experiments

[0314] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[0315] At the same time, the order in which the analysis results are presented and the amount of detailed explanation are adjusted based on the user's emotions.

[0316] 8. Providing the results

[0317] The server provides the user with generated summaries, image analysis results, mathematical formula interpretations, and code for reproduction experiments.

[0318] Users receive information adjusted through emotion recognition and can utilize the analysis results in a format that suits their own emotions, thereby facilitating the smooth progress of replication experiments.

[0319] The following describes the processing flow.

[0320] Step 1:

[0321] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0322] Step 2:

[0323] The server receives the PDF file and temporarily saves it to storage. For example, the save location is set to / tmp / uploaded_paper.pdf, and the file path is checked when passing it to the next process.

[0324] Step 3:

[0325] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Text is extracted from each page and combined into a single continuous text file. Images are saved to a specific directory, and mathematical formulas are output as images.

[0326] Step 4:

[0327] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. For example, text can be sent using an API, and the results can be saved in text format.

[0328] python

[0329] import openai

[0330] openai.api_key = 'your-api-key'

[0331] response = openai.Completion.create(

[0332] engine="text-davinci-003",

[0333] prompt="Analyze and summarize the following text:\n" + all_text,

[0334] max_tokens=500

[0335] )

[0336] summary = response.choices[0].text

[0337] Preservation process

[0338] Step 5:

[0339] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts graph elements and features from each image and saves information such as model parameters and activation functions in text format.

[0340] python

[0341] from yolov5 import YOLOv5

[0342] yolo = YOLOv5("path / to / yolov5 / weights")

[0343] for img_path in image_save_paths:

[0344] results = yolo.predict(img_path)

[0345] parsed_results = results.pandas().xyxy Save process

[0346] Step 6:

[0347] The server extracts mathematical formulas and analyzes them using a mathematical formula analysis tool (e.g., Mathpix). OCR technology is used to convert the mathematical formula images into text format, and the resulting text data is saved in a format that can be understood.

[0348] python

[0349] import pytesseract

[0350] Import Image from PIL

[0351] math_images = [] Assumes that the images of mathematical formulas have been identified and saved beforehand.

[0352] math_texts = [pytesseract.image_to_string(Image.open(img_path), config="math_configuration") for img_path in math_images]

[0353] Interpretation and storage

[0354] Step 7:

[0355] The server uses an emotion engine to recognize the user's emotions in real time. Emotional data is acquired based, for example, on user input, interactions, and video analysis from the webcam. Once emotional data is acquired, the analysis method and the way results are provided are adjusted accordingly.

[0356] python

[0357] import emotion_recognition

[0358] emotions = emotion_recognition.recognize_webcam()

[0359] Preservation process

[0360] Step 8:

[0361] The server integrates the analysis results of text, images, and mathematical formulas to generate the program code necessary for reproduction experiments. It combines the results of natural language processing models, image analysis, and mathematical formula analysis to generate Python code that enables reproduction experiments. In addition, it adds guides and explanations tailored to the user's emotions.

[0362] Step 9:

[0363] The server provides the user with the generated analysis results and program code. The method of delivery is adjusted according to the user's preferences; for example, a detailed step-by-step tutorial may be provided. The user then uses this to perform reproduction experiments.

[0364] (Example 2)

[0365] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the smart device 14 as the "terminal".

[0366] In modern research, it is crucial to quickly and accurately understand the content of research papers and to conduct replication experiments. However, manually analyzing a vast number of papers and extracting the information necessary for summarization and replication experiments is an extremely time-consuming task. Furthermore, providing information that disregards the user's feelings is ineffective in understanding and utilizing the analysis results. This makes it difficult for researchers to conduct replication experiments efficiently, leading to a slowdown in research progress.

[0367] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0368] In this invention, the server includes means for receiving electronic documents from a user, means for extracting document information, image information, and mathematical formula information from the electronic documents, means for using a natural language processing model to analyze the extracted document information, means for using an image analysis model to analyze the extracted image information, means for using a mathematical formula analysis tool to analyze the extracted mathematical formula information, means for integrating the analysis results and generating program code necessary for experimental reproduction, means for recognizing the user's emotions and adjusting the method and order of presentation of the analysis results, and means for providing the user with the analysis results and the generated program code. This makes it possible to quickly and accurately analyze the content of a paper and efficiently provide the information necessary for reproduction experiments. By providing information in accordance with the user's emotions, it is expected that the understanding of the analysis results will be enhanced and the reproduction experiments will proceed smoothly.

[0369] A "user" is an individual or organization that uses the system to upload electronic documents and receives analysis results and program code.

[0370] An "electronic document" refers to documents such as academic papers and research materials that are saved in PDF format or other digital formats.

[0371] "Document information" refers to text data extracted from electronic documents.

[0372] "Image information" refers to image data extracted from electronic documents.

[0373] "Mathematical formula information" refers to mathematical formula data extracted from electronic documents.

[0374] A "natural language processing model" is a machine learning model or algorithm used to analyze extracted document information and generate summaries and key points.

[0375] An "image analysis model" is a machine learning model or algorithm used to analyze extracted image information and extract the constituent elements and features of a figure or chart.

[0376] A "mathematical formula analysis tool" is software or algorithms that analyze extracted mathematical formula information and generate the meaning and application examples of the formulas in text format.

[0377] "Analysis results" refer to the results obtained using natural language processing models, image analysis models, and mathematical analysis tools.

[0378] "Program code" refers to the source code generated based on the analysis results, used to execute the reproduction experiment.

[0379] "Emotion recognition" refers to the process and technology of analyzing a user's emotions and adjusting the way results are presented and the order in which they are presented.

[0380] This invention is a system that analyzes electronic documents and comprehensively understands document information, image information, and mathematical formula information. This system allows users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system primarily operates on the server side, analyzing electronic documents uploaded by users and providing the results.

[0381] The server performs the following process: First, the user uploads an electronic document via the web interface. Then, the server receives this electronic document and temporarily stores it in storage. A possible specific storage location might be / tmp / uploaded_document.pdf.

[0382] Next, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information. Each page is analyzed separately, and the text is saved as a single continuous document, while images and mathematical formulas are saved in separate directories (e.g., / tmp / images / ).

[0383] The extracted document information is analyzed using a natural language processing model (e.g., GPT-4). This analysis generates a summary and key points of the paper, and the results are saved in text format (e.g., / tmp / summary.txt). Similarly, the extracted image information is analyzed using an image analysis model (e.g., YOLO, CNN), and the information obtained from the images is saved in text format (e.g., / tmp / image_analysis.txt). In addition, mathematical formula information is analyzed using a mathematical formula analysis tool (e.g., Mathpix), and the meaning and application examples of the formulas are described in natural language and saved as text data (e.g., / tmp / formula_analysis.txt).

[0384] Furthermore, the server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed in real time through analysis of the user's past interactions and video from the webcam (if permitted). It determines whether the user is showing anxiety or interest and stores the results in an internal data structure.

[0385] These analysis results are integrated to generate the program code necessary for reproduction experiments. The generated code includes the model definition, data preprocessing, and training procedures. For example, the generated code is saved in / tmp / reproduction_code.py. During this process, the server adjusts the order in which results are presented and the level of detail in the explanations based on the user's emotions. For example, if the user is feeling anxious, it provides detailed explanations and step-by-step guidance.

[0386] Finally, the server provides the user with the generated analysis results and program code. The user can download these from the web interface and use them as reference information for conducting reproduction experiments.

[0387] As a concrete example of operation, the following prompt statements are possible:

[0388] "I have uploaded a PDF of a paper on a new gradient descent method. Please analyze the contents of this PDF. Generate a summary of the text, images, and mathematical formulas, and then create Python code for a replication experiment based on these. Also, please adjust the presentation method of the analysis results based on my interests and concerns."

[0389] As described above, this system helps advance research activities by reducing the burden on users and efficiently providing analysis results.

[0390] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0391] Step 1:

[0392] Users upload electronic documents (PDF files) to the server via a web interface.

[0393] Specifically, the user opens a browser and clicks the file upload button. They select a PDF file from the file selection dialog and click the "Upload" button.

[0394] Input: PDF file uploaded by the user

[0395] Output: PDF file sent to the server

[0396] Step 2:

[0397] The server receives the uploaded PDF file and temporarily stores it in storage.

[0398] Specifically, the file is saved to / tmp / uploaded_document.pdf.

[0399] Input: PDF file sent by the user to the server

[0400] Output: PDF file saved in the specified directory on the server.

[0401] Step 3:

[0402] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information.

[0403] Specifically, each page is analyzed, the text is saved as a continuous document, and images and formulas are saved in a separate directory (e.g., / tmp / images / ).

[0404] Input: Saved PDF file

[0405] Output: A dataset of extracted text, images, and mathematical formulas.

[0406] Step 4:

[0407] The server uses a natural language processing model (e.g., GPT-4) to analyze the extracted document information.

[0408] Specifically, the extracted text is input into a text analysis model, which then generates a summary and key points. The generated analysis results are saved in text format to / tmp / summary.txt.

[0409] Input: Extracted text data

[0410] Output: A text file containing summaries and key points.

[0411] Step 5:

[0412] The server analyzes image information using an image analysis model (e.g., YOLO, CNN).

[0413] Specifically, the image analysis model is run using each image file as input, and the information obtained from the analysis is saved to / tmp / image_analysis.txt.

[0414] Input: Extracted image data

[0415] Output: Text file containing image analysis results

[0416] Step 6:

[0417] The server uses a mathematical analysis tool (e.g., Mathpix) to analyze the mathematical information.

[0418] Specifically, the tool takes an image of a mathematical formula as input and converts it into text format. It then describes the meaning and application examples of the formula in natural language and saves them to / tmp / formula_analysis.txt.

[0419] Input: Extracted formula data

[0420] Output: A text file containing the results of the mathematical formula analysis.

[0421] Step 7:

[0422] The server uses an emotion engine to recognize the user's emotions.

[0423] Specifically, the system analyzes user input, interaction data, and, if permitted, webcam footage to determine whether the user is showing anxiety or interest. The results are stored in an internal data structure.

[0424] Input: User interaction data and video data

[0425] Output: Data on the user's emotional state

[0426] Step 8:

[0427] The server integrates the analysis results so far and generates the program code necessary for the reproduction experiment.

[0428] Specifically, the system combines the results of text analysis, image analysis, and mathematical analysis into a single file. It then incorporates a summary of the natural language processing model to generate Python code that allows for reproduction experiments. The generated code is saved in ` / tmp / reproduction_code.py`. Furthermore, the system adjusts the presentation order and level of detail of the results based on the user's emotions.

[0429] Input: Text analysis results, image analysis results, mathematical formula analysis results, user emotional state data

[0430] Output: Python code for reproduction experiment

[0431] Step 9:

[0432] The server provides the user with the generated analysis results and program code.

[0433] Specifically, the analysis results and code are presented as download links on a web interface. Users download them and perform reproduction experiments.

[0434] Input: Generated analysis results and program code

[0435] Output: Analysis results and program code provided to the user.

[0436] (Application Example 2)

[0437] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as a "server" and the smart device 14 as a "terminal".

[0438] Currently, it is difficult for workers in logistics centers to quickly understand information about new technologies and algorithms and perform tasks efficiently based on that understanding. Furthermore, because there is no system in place to understand and appropriately respond to workers' emotions, anxiety and stress increase when the learning curve is steep, leading to decreased work efficiency. Therefore, there is a need to develop a system that can quickly and accurately analyze the content of research papers and adjust the method of information delivery according to the workers' emotions.

[0439] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for receiving a PDF of a paper from a user, means for extracting text, images, and mathematical formulas from the PDF, means for using a natural language processing model to analyze the extracted text, means for using an image analysis model to analyze the extracted images, means for using a mathematical formula analysis tool to analyze the extracted mathematical formulas, means for recognizing the user's emotions, means for integrating the analysis results and generating program code necessary for reproduction experiments, means for adjusting the method and presentation order of the analysis results based on the user's emotions, and means for providing the user with the analysis results and the generated program code. This enables an assist system that allows workers in a logistics center to quickly understand new technical information and perform work efficiently.

[0440] A "user" is a person or entity that uses the system to upload a PDF of a research paper and receive the analysis results.

[0441] A "paper" is a document that describes the results of academic research or studies in PDF format.

[0442] "PDF" stands for Portable Document Format, a file format that can contain text, images, and mathematical formulas in a document.

[0443] "Text" refers to the written information contained in a PDF, including sentences, explanations, and comments.

[0444] An "image" is an element that represents visual information such as graphs, diagrams, and photographs included within a PDF document.

[0445] "Mathematical formulas" refer to mathematical expressions written within a PDF document, including calculation formulas and equations.

[0446] A "natural language processing model" is an artificial intelligence technology used to analyze, summarize, and interpret text, such as a model that understands the meaning of extracted text.

[0447] An "image analysis model" is an artificial intelligence technology used to analyze images; for example, it is a model used to extract and understand information contained in an image.

[0448] A "mathematical formula analysis tool" is a tool used to analyze mathematical formulas and understand their meaning and application examples.

[0449] "Emotion recognition" is a technology that analyzes and identifies a user's emotions, and is about understanding the user's state of mind.

[0450] "Analysis results" refer to information and content generated based on the analysis of text, images, and mathematical formulas extracted from the PDF.

[0451] "Program code" refers to computer language code that describes the procedures and operations necessary to carry out a reproduction experiment.

[0452] "Means for adjusting the method of provision and presentation order" refers to a function that changes the way analysis results are displayed and the order in which they are presented based on the user's emotions.

[0453] To implement this invention, it is necessary to build a system that includes the following steps. Specific hardware includes a head-mounted display (HMD) for user use, a server, and a webcam. Software used may include PDF parsing libraries (e.g., PyMuPDF or PDFMiner), OCR software (e.g., Pytesseract), natural language processing models (e.g., Hugging Face Transformers), emotion recognition engines (e.g., EmotionRecognizer), and image analysis models (e.g., YOLO or CNN).

[0454] First, the user uploads the PDF of their paper to the server via the HMD (Head-Mounted Display). The server receives the PDF file and temporarily stores it in storage. This file is later broken down into text, images, and mathematical formulas using a PDF analysis library. The text is extracted using OCR software, and the images are saved in a separate directory. The mathematical formulas are saved as images and analyzed using a mathematical formula analysis tool.

[0455] Next, the server uses a natural language processing model to analyze the extracted text and generate a summary and key points of the paper. An image analysis model is used to extract elements and features of figures and tables from images, and the resulting information is saved in text format. A mathematical formula analysis tool is used to obtain text information from mathematical formula images, explaining the meaning and application examples of the formulas.

[0456] The server also uses a webcam to recognize the user's emotions in real time. The emotion recognition engine determines emotions from the user's input, interactions, and facial expressions, and evaluates whether the user is showing anxiety or interest.

[0457] The server integrates these analysis results and automatically generates the Python code necessary for replication experiments. It adjusts the presentation order and detailed explanations of the analysis results based on the user's emotional state. For example, if the user is feeling anxious, it provides a detailed step-by-step guide to help them confidently replicate the experiment.

[0458] Finally, the generated analysis results and program code are provided to the user via the HMD. The user can then use this as a reference to perform reproduction experiments.

[0459] As a concrete example, when a worker at a logistics center uploads a paper on a new cargo placement algorithm, the analysis results are displayed on a head-mounted display. If the worker has any questions or concerns, an emotion recognition engine detects this and automatically provides a detailed step-by-step guide.

[0460] Examples of prompts include: "Analyze the key points of the latest research paper on logistics centers. A detailed explanation of specific cargo placement algorithms is required. Also, suggest ways to alleviate worker anxiety."

[0461] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0462] Step 1:

[0463] The user uploads a PDF of their paper to the server via a head-mounted display. The input is the PDF file selected by the user, and the output is this PDF file being saved on the server. The server temporarily stores the file in storage and passes it on to the next process.

[0464] Step 2:

[0465] The server parses a PDF file and extracts text, images, and mathematical formulas. The input is a PDF file stored on the server, and the output is the extracted text, images, and mathematical formulas. Specifically, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to process each page, concatenating the text into a single continuous text file, saving the images in a separate directory, and saving the mathematical formulas as images.

[0466] Step 3:

[0467] The server uses a natural language processing model to analyze the extracted text. The input is the extracted text, and the output is a summarized text or key points. Specifically, the server uses a natural language processing model (e.g., Hugging Face Transformers) to summarize the main sections of a paper and extract the core of a new method.

[0468] Step 4:

[0469] The server uses an image analysis model to analyze the extracted images. The input is the extracted images, and the output is text data of the constituent elements and features derived from the images. Specifically, the server uses an image analysis model (e.g., YOLO or CNN) to analyze the components of graphs and charts from the images and extract parameters and activation functions for each layer.

[0470] Step 5:

[0471] The server uses a mathematical expression analysis tool to analyze the extracted mathematical formulas. The input is an image of the formula, and the output is an interpretation of the formula in text format. Specifically, the server uses a mathematical expression analysis tool (e.g., Mathpix) to convert the image of the formula into text format and explains the meaning and application examples of the formula.

[0472] Step 6:

[0473] The server uses a webcam to recognize the user's emotions. The input is the user's video captured by the webcam, and the output is the estimated emotion of the user. Specifically, the server uses an emotion recognition engine (e.g., EmotionRecognizer) to determine the emotion in real time from the user's facial expressions and interactions.

[0474] Step 7:

[0475] The server integrates the analysis results to date and generates the program code necessary for the replication experiment. Inputs include analysis results of text, images, and mathematical formulas, as well as user sentiment information. Outputs are Python code required for the replication experiment. Specifically, the server uses a generative AI model to integrate the results and automatically generates program code that enables the replication experiment.

[0476] Step 8:

[0477] The server adjusts the method and order in which analysis results are presented based on the user's emotions. Input is the user's emotion information and generated program code, while output is the analysis results and program code formatted according to the emotion. Specifically, the server adds detailed explanations and step-by-step guides based on the emotion recognition results.

[0478] Step 9:

[0479] The server provides the user with the generated analysis results and program code. The input is the adjusted analysis results and program code, and the output is the result displayed on the user's head-mounted display. Specifically, the server allows the user to download or view the analysis results and code via a web interface.

[0480] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0481] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0482] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the smart device 14.

[0483] [Second Embodiment]

[0484] Figure 3 shows an example of the configuration of the data processing system 210 according to the second embodiment.

[0485] As shown in Figure 3, the data processing system 210 includes a data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0486] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0487] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication interface 44. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, and camera 42 are also connected to the bus 52.

[0488] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[0489] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[0490] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[0491] Figure 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Figure 4, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[0492] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0493] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0494] In the smart glasses 214, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0495] Next, the identification processing performed by the identification processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0496] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. This system operates on a server side and analyzes PDFs uploaded by users, providing the results.

[0497] Program processing flow

[0498] 1. Receiving the PDF

[0499] Users upload their research papers as PDFs to the server via a web interface. This serves as the entry point to the system.

[0500] 2. Data Extraction

[0501] The server saves the received PDF file to a specific directory. Next, a PDF parsing library is used to extract text, images, and mathematical formulas from the PDF.

[0502] The text is read page by page, combined, and saved as a single continuous text file. Images are saved in a specific folder, and mathematical formulas are converted into text format using OCR technology.

[0503] 3. Text Analysis

[0504] The server uses a natural language processing model to analyze the extracted text. This generates a summary and key points of the paper.

[0505] 4. Image Analysis

[0506] The server uses an image analysis model to analyze images within the paper. It retrieves data points from graphs and figures contained within the images, as well as their interpretations, in text format.

[0507] 5. Formula Analysis

[0508] The server uses a mathematical analysis tool to analyze the extracted mathematical formulas. The analysis results are saved as text data, including interpretations of the formulas' meanings and applications.

[0509] 6. Support for Integration and Reproducibility Experiments

[0510] The server integrates the analysis results and generates code for replication experiments. This allows users to obtain specific procedures and program code to reproduce the experiment based on the content of the paper.

[0511] 7. Providing the results

[0512] The server returns the generated analysis results and program code to the user. The user can then use this to gain a deeper understanding of the paper's content and smoothly carry out replication experiments.

[0513] Specific example

[0514] As a concrete example, we can consider the following scenario.

[0515] 1. Receiving the PDF

[0516] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[0517] 2. Data Extraction

[0518] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[0519] 3. Text Analysis

[0520] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[0521] 4. Image Analysis

[0522] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[0523] 5. Formula Analysis

[0524] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0525] 6. Support for Integration and Reproducibility Experiments

[0526] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[0527] 7. Providing the results

[0528] The server provides the user with a generated summary, image analysis results, mathematical formula interpretations, and code for replication experiments. The user can then use this information to reproduce the paper's experiments in their own environment.

[0529] As described above, the system of the present invention enables users to deeply understand the content of research papers and efficiently conduct reproduction experiments.

[0530] The following describes the processing flow.

[0531] Step 1:

[0532] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0533] Step 2:

[0534] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[0535] Step 3:

[0536] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[0537] Step 4:

[0538] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. The analysis results are saved in text format and used for subsequent processing.

[0539] Step 5:

[0540] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts the constituent elements and features of each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[0541] Step 6:

[0542] The server extracts mathematical formulas and analyzes them using a mathematical analysis tool (e.g., Mathpix). The formula images are converted into text format, and the meaning and application examples of the formulas are described in natural language. This allows the content of the formulas to be saved as text data.

[0543] Step 7:

[0544] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[0545] Step 8:

[0546] The server provides the user with the generated analysis results and program code. The user downloads this from the web interface and uses it as a reference to conduct reproduction experiments.

[0547] (Example 1)

[0548] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the smart glasses 214 will be referred to as the "terminal."

[0549] Conventional technologies required advanced expertise and considerable time to comprehensively analyze the content of papers and documents and facilitate reproduction experiments. Furthermore, the process of integrating results using multiple analysis methods and tools was complex and burdensome for users. Therefore, the present invention aims to provide a system that automates document analysis and preparation for reproduction experiments, enabling efficient and accurate execution.

[0550] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0551] In this invention, the server includes means for receiving a document file from a user, means for extracting character data, figures, and formulas from the file, means for using a natural language processing model to analyze the extracted character data, means for using an image analysis model to analyze the extracted figures, means for using an analysis tool to analyze the extracted formulas, means for integrating the analysis results and generating a program necessary for reproduction verification, and means for providing the user with the analysis results and the generated program. This makes it possible to comprehensively analyze the contents of a document and to perform reproduction experiments efficiently and accurately.

[0552] A "user" refers to an entity that uploads document files to the server and receives the analysis results or programs.

[0553] "Document" refers to electronic files containing text information such as papers and reports, including formats such as PDF.

[0554] A "file" refers to digital data in which the contents of a document are stored.

[0555] "Character data" refers to the text information contained within a document.

[0556] "Graphics" refers to visual information such as images, graphs, and charts included within a document.

[0557] "Formula" refers to mathematical formulas or mathematical expressions included in a document.

[0558] "Extraction" refers to the process of taking specific data (text data, graphics, formulas) out of a file.

[0559] A "natural language processing model" refers to a machine learning model or algorithm used to analyze extracted text data and extract important information.

[0560] An "image analysis model" refers to a machine learning model or algorithm that analyzes extracted graphic data and converts visual information into text format.

[0561] An "analysis tool" refers to software or algorithms used to analyze extracted formulas and understand their meaning.

[0562] "Integration" refers to the process of combining analysis results into a single format and ensuring consistency.

[0563] "Reproduction verification" refers to the process of reproducing experiments or simulations based on the content of a document.

[0564] A "program" refers to a set of code and procedures necessary for reproducing and verifying a problem.

[0565] "Results" refers to the information obtained from the analysis of the document.

[0566] This invention is a document analysis and reproduction experiment support system, primarily configured to operate on the server side. The specific implementation methods for each step are described below.

[0567] Receiving document files

[0568] Users upload document files to the server using a web interface. These document files include PDFs and other formats containing text information such as papers and reports. The server saves the received document files to a specific directory (e.g., / tmp / pdf_uploads / ).

[0569] Data extraction

[0570] The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text data, shapes, and formulas from the file. The specific steps are as follows:

[0571] Text data: Extracts text from each page of a document and saves it as a series of text data.

[0572] Shapes: Extracts images, graphs, and charts from a document and saves them, for example, to a folder such as / tmp / pdf_uploads / images / .

[0573] Formula: Convert mathematical formulas to text format using an OCR library (e.g., Mathpix) and save them.

[0574] Data analysis

[0575] Analysis of character data

[0576] The server uses natural language processing models (e.g., BERT, GPT) to analyze text data and extract summaries and key points from papers. This allows the server to generate summaries that enable users to quickly understand the content of documents.

[0577] Analysis of geometric shapes

[0578] The server analyzes the shapes using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format. The analysis results include, for example, interpretations of the model's architecture diagram and experimental results graphs.

[0579] Analysis of the formula

[0580] The server uses a mathematical analysis tool (e.g., SymPy) to analyze mathematical formulas and generate interpretations of their meaning and applications. The analysis results are saved as text data explaining the meaning of the formulas.

[0581] Support for integration and replication experiments

[0582] The server integrates these analysis results and generates a program for replication experiments. This program is provided as Python code that includes model definition, data preprocessing, and training procedures. Users can use this program to replicate the experiment in their own environment.

[0583] Providing results

[0584] The server returns the generated analysis results and program code to the user. The analysis results include a summary, image analysis results, interpretation of mathematical formulas, and code for reproduction experiments, enabling the user to gain a deeper understanding of the document's content and efficiently conduct reproduction experiments.

[0585] Specific example

[0586] As a concrete example, consider a case where a user uploads a paper on a new learning method for AI models. The server can perform the analysis through the following steps:

[0587] The server extracts 40 pages of text, 5 images, and 3 mathematical formulas, and analyzes each of them.

[0588] The server uses a natural language processing model to generate a summary that reads, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[0589] The server uses an image analysis model to analyze the image, concluding that "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer."

[0590] The server uses a mathematical analysis tool to interpret Equation 1 as "showing the loss function of the proposed method, which minimizes the squared error."

[0591] The server generates Python code for reproduction experiments and provides it to the user.

[0592] Examples of prompt statements

[0593] "Analyze a paper on a new AI model learning method, extract key points, interpret image content and mathematical formulas, and generate code for replication experiments."

[0594] "Please analyze the following PDF paper and provide the summary, data points for the graphs, meaning of the formulas, and experimental reproduction procedures. PDF link: [link]"

[0595] As described above, the document analysis system of the present invention can provide an effective means for users to deeply understand the content of documents and efficiently conduct reproduction experiments.

[0596] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0597] Step 1:

[0598] PDF reception

[0599] Input: The PDF file to be uploaded by the user.

[0600] Specific operation: On the web interface, the user clicks the "Select File" button, selects the desired PDF file from their local disk, and clicks the upload button.

[0601] Data processing: The server saves the received PDF files to a specific directory (e.g., / tmp / pdf_uploads / ).

[0602] Output: PDF file saved in / tmp / pdf_uploads / .

[0603] Step 2:

[0604] Data extraction

[0605] Input: Saved PDF file.

[0606] Specific operation: The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text, images, and mathematical formulas from PDF files.

[0607] Text extraction: Reads text from each page and saves it as a series of text files (e.g., / tmp / pdf_uploads / extracted_text.txt).

[0608] Image extraction: Identifies images within the PDF and saves them to a specific folder (e.g., / tmp / pdf_uploads / images / ).

[0609] Formula extraction: Use an OCR library (e.g., Mathpix) to convert formulas in the PDF into text format and save it (e.g., / tmp / pdf_uploads / extracted_equations.txt).

[0610] Data processing: Extraction of text, images, and mathematical formulas using a PDF analysis library.

[0611] Output: Text data ( / tmp / pdf_uploads / extracted_text.txt), image data ( / tmp / pdf_uploads / images / ), mathematical formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[0612] Step 3:

[0613] Text analysis

[0614] Input: Extracted text data ( / tmp / pdf_uploads / extracted_text.txt).

[0615] Specific operation: The server uses a natural language processing model (e.g., BERT, GPT) to analyze the text data and extract summaries and key points.

[0616] Section Identification: Analyzes text data to identify sections of a paper (e.g., Introduction, Methods, Results, Discussion, etc.).

[0617] Summary generation: Generate summaries for each section and extract the core elements of the new learning method.

[0618] Data processing: Text analysis and summarization using natural language processing models.

[0619] Output: Summarized text data.

[0620] Step 4:

[0621] Image analysis

[0622] Input: Extracted image data ( / tmp / pdf_uploads / images / ).

[0623] Specific operation: The server analyzes images using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format.

[0624] Graph and chart analysis: Extracting data points and interpretations from images (e.g., "Figure 1 shows a 3-layer convolutional neural network, with ReLU used in each layer").

[0625] Data processing: Analysis of image data using image analysis models.

[0626] Output: Text data of the image analysis results.

[0627] Step 5:

[0628] Formula analysis

[0629] Input: Extracted formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[0630] Specific operation: The server uses a mathematical expression analysis tool (e.g., SymPy) to analyze the formula and generate an interpretation of its meaning and application.

[0631] Mathematical analysis: Analyze the mathematical formula and interpret it as "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0632] Data calculation: Analysis of mathematical formulas using mathematical analysis tools.

[0633] Output: Text data of the mathematical analysis results.

[0634] Step 6:

[0635] Support for integration and replication experiments

[0636] Input: Text analysis results, image analysis results, mathematical formula analysis results.

[0637] Specific operation: The server integrates these analysis results and generates a program for reproduction experiments.

[0638] Integration of analysis results: Organize the analysis results and compile them into a single document (e.g., Markdown format).

[0639] Program generation: Generates Python code for replication experiments, describing the model definition, data preprocessing, and training procedures.

[0640] Data processing: Integration of analysis results and generation of program code.

[0641] Output: Integrated analysis results document and Python code.

[0642] Step 7:

[0643] Providing results

[0644] Input: Integrated analysis results document and Python code.

[0645] Specific operation: The server compresses the generated analysis results and program code into a zip file and provides a download link for the user.

[0646] Output: A download link that the user can access.

[0647] The above describes the specific operation and input / output of each processing step in this system.

[0648] (Application Example 1)

[0649] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server," and the smart glasses 214 will be referred to as the "terminal."

[0650] When new production technologies or methods are proposed within a factory, actually reproducing and verifying them requires time and effort. Furthermore, understanding technical documents such as research papers requires specialized knowledge, and appropriately interpreting their content and reflecting it in experiments is difficult. As a result, the introduction of new technologies is often delayed, hindering improvements in production efficiency and quality.

[0651] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[0652] In this invention, the server includes means for receiving a PDF of a technical document from a user; means for extracting information, images, and mathematical representations from the PDF; means for using a natural language processing model to analyze the extracted information; means for using an image processing model to analyze the extracted images; means for using a mathematical expression analysis tool to analyze the extracted mathematical representations; means for integrating the analysis results and generating program code and automated experimental procedures necessary for reproduction experiments; and means for providing the analysis results and generated program code to the user. This makes it possible to quickly analyze new technologies described in technical documents within a factory and to automatically perform reproduction experiments.

[0653] A "user" is an individual or organization that operates the system to analyze technical documents or conduct reproduction experiments.

[0654] "Technical documents" refer to documents containing content related to science, technology, or engineering, such as research papers and technical reports.

[0655] "PDF" is an abbreviation for Portable Document Format, a file format for electronically displaying the contents of a document.

[0656] "Information" refers collectively to the text, images, and mathematical formulas included in technical documents.

[0657] "Images" refer to visual data such as diagrams, graphs, and photographs included within technical documents.

[0658] "Mathematical expression" refers to mathematical formulas and statistical descriptions included in technical documents.

[0659] A "natural language processing model" is a computational model used to analyze text data and understand its meaning and structure.

[0660] An "image processing model" is a computational model used to analyze image data and extract its content and features.

[0661] A "mathematical formula analysis tool" is software used to analyze mathematical formulas and understand their meaning and structure.

[0662] "Analysis results" refers to the collective data obtained by natural language processing models, image processing models, and mathematical analysis tools.

[0663] "Program code" refers to a computer program that describes the procedures and processes necessary to carry out a reproduction experiment.

[0664] An "automated experimental procedure" is a specific set of steps generated based on analysis results, which are used to perform a reproducible experiment using automated equipment.

[0665] A "system" is a general term for a complex set of devices or software that automates everything from the analysis of technical documents to reproduction experiments.

[0666] This invention provides a system that analyzes PDF technical documents and comprehensively analyzes text, images, and mathematical representations to facilitate users in performing replication experiments. This system includes a paper decoding system, a server, a cloud server, and a factory robot.

[0667] System Overview

[0668] Program processing

[0669] The server first receives technical document PDFs from users. The received PDFs are saved in a specific directory. Next, a PDF parsing library (e.g., PyPDF2) is used to extract text, images, and mathematical representations from the PDFs. The extracted text is analyzed using a natural language processing model (e.g., spacy) to generate summaries and key points of the papers. Images are analyzed using an image processing model (e.g., pytesseract) to convert the included data and graphs into text format. Mathematical representations are analyzed using a mathematical expression analysis tool (e.g., regular expressions) to explain their meaning.

[0670] The analysis results are integrated to generate the program code and automated experimental procedures necessary for reproduction experiments. This generated code is provided to the user. Furthermore, the system provides procedures and tools for automatically performing reproduction experiments based on the analysis results.

[0671] Hardware and software used

[0672] Hardware:

[0673] Factory robots (e.g., KUKA robots, ABB robots)

[0674] Cloud servers (AWS, Google Cloud, etc.)

[0675] software:

[0676] PDF analysis: PyPDF2

[0677] Natural Language Processing: Spacy

[0678] Image analysis: pytesseract

[0679] Formula extraction: Regular expressions (re)

[0680] Specific example

[0681] As a concrete example, consider a scenario in which a technical document on a new, highly efficient production technology is introduced to a factory. First, the factory manager uploads the PDF of the technical document to a factory robot using a smartphone. Analysis is performed on a cloud server, and the following results are generated.

[0682] Summary: "This technical document proposes a new technology that achieves a 20% improvement in efficiency compared to conventional technology."

[0683] Image analysis: "Figure 1 shows a flowchart of the new technology, and the main steps are as follows..."

[0684] Mathematical analysis: "Equation 1 shows the optimization parameters for improving efficiency."

[0685] Based on these analysis results, detailed program code necessary for reproduction experiments is generated. This generated code is provided to the factory robot, which automatically executes the reproduction experiment. This process allows for rapid verification of the effectiveness of new technologies and supports their implementation within the factory.

[0686] Example of a prompt

[0687] Examples of specific prompt messages are as follows:

[0688] "Please write a program for industrial robots to analyze papers on new production technologies and automate the replication experiments."

[0689] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0690] Step 1:

[0691] PDF reception

[0692] The user uploads a PDF file of a technical document. This PDF is received by the server and saved to a specific directory.

[0693] Input: PDF file of a technical document.

[0694] Output: PDF file saved in the server directory.

[0695] Specific operation: When a user selects a PDF from the web interface and presses the upload button, the server saves the file to the specified directory.

[0696] Step 2:

[0697] Data extraction

[0698] The server uses the PDF parsing library (PyPDF2) to extract text, images, and mathematical representations from PDF files.

[0699] Input: Saved PDF file.

[0700] Output: Extracted text files, a list of image files, and a list of mathematical representations.

[0701] Specific operation: The server opens the PDF file, extracts text from each page, saves the image files, and adds the mathematical representations to a list.

[0702] Step 3:

[0703] Text analysis

[0704] The server uses a natural language processing model (SPacy) to analyze the extracted text.

[0705] Input: Extracted text file.

[0706] Output: Text summary and key points.

[0707] Specific operation: The server inputs a text file into a natural language processing model and extracts a summary and important sections.

[0708] Step 4:

[0709] Image analysis

[0710] The server uses an image processing model (pytesseract) to analyze the extracted images.

[0711] Input: List of extracted image files.

[0712] Output: Text data extracted from the image.

[0713] Specific operation: The server sequentially inputs each image file into the image processing model and extracts the text data.

[0714] Step 5:

[0715] Formula analysis

[0716] The server uses a mathematical expression analysis tool (regular expressions) to analyze the extracted mathematical representation.

[0717] Input: A list of mathematical expressions.

[0718] Output: Interpretation and explanation of the meaning of the formula.

[0719] Specific operation: The server inputs the list of mathematical formulas into a regular expression tool and generates text explaining the meaning of each formula.

[0720] Step 6:

[0721] Support for integration and replication experiments

[0722] The server integrates the analysis results and generates the program code and automated experimental procedures necessary for reproduction experiments.

[0723] Input: Text summaries, text data extracted from images, and interpretation of mathematical formulas.

[0724] Output: Integrated analysis results and program code for replication experiments.

[0725] Specific operation: The server integrates all data and generates Python code for reproduction experiments based on the analysis results.

[0726] Step 7:

[0727] Providing results

[0728] The server provides the user with the generated analysis results and program code.

[0729] Input: Integrated analysis results and program code for replication experiments.

[0730] Output: Analysis results and program code provided to the user.

[0731] Specific operation: The server formats the analysis results and code and provides them to the user as a download link. It also sends the analysis results via email or other methods as needed.

[0732] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0733] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system operates on a server side, analyzing PDFs uploaded by users and providing the results.

[0734] Program processing flow

[0735] 1. Receiving the PDF

[0736] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0737] 2. Data Extraction

[0738] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[0739] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[0740] 3. Text Analysis

[0741] The server utilizes natural language processing models (e.g., GPT-4) to analyze the extracted text. This generates summaries and key points of the papers. The analysis results are saved in text format and used for subsequent processing.

[0742] 4. Image Analysis

[0743] The server performs the analysis using an image analysis model (e.g., YOLO, CNN). It extracts the components and features of the figures and tables from each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[0744] 5. Formula Analysis

[0745] The server uses a mathematical formula analysis tool (e.g., Mathpix) to perform the analysis. It converts the mathematical formula image into text format and describes the meaning and application examples of the formula in natural language. This allows the content of the formula to be saved as text data.

[0746] 6. Emotion recognition

[0747] The server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed through user input, interaction, and video analysis from the webcam. For example, it can determine in real time whether the user is showing anxiety or interest.

[0748] 7. Support for Integration and Reproducibility Experiments

[0749] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[0750] Furthermore, the method of providing analysis results will be adjusted based on the user's emotions. For example, if a user is feeling anxious, more detailed explanations and step-by-step guides will be provided.

[0751] 8. Providing results

[0752] The server provides the user with the generated analysis results and program code. The order and method of presenting the results are adjusted according to the user's mood. For example, if the user appears interested, important results can be presented first to maintain their attention.

[0753] Users download this from the web interface and use it as a reference for conducting reproduction experiments. The reproduction experiment process proceeds based on the provided code and explanations.

[0754] Specific example

[0755] As a concrete example, we can consider the following scenario.

[0756] 1. Receiving the PDF

[0757] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[0758] 2. Data Extraction

[0759] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[0760] 3. Text Analysis

[0761] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[0762] 4. Image Analysis

[0763] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[0764] 5. Formula Analysis

[0765] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0766] 6. Emotion recognition

[0767] The server recognizes the user's emotions and determines whether the user is feeling anxious about the analysis results. For example, if a user has asked many questions about past analysis results, the server will determine that the user is feeling anxious.

[0768] 7. Support for Integration and Reproducibility Experiments

[0769] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[0770] At the same time, the order in which the analysis results are presented and the amount of detailed explanation are adjusted based on the user's emotions.

[0771] 8. Providing results

[0772] The server provides the user with generated summaries, image analysis results, mathematical formula interpretations, and code for reproduction experiments.

[0773] Users receive information adjusted through emotion recognition and can utilize the analysis results in a format that suits their own emotions, thereby facilitating the smooth progress of replication experiments.

[0774] The following describes the processing flow.

[0775] Step 1:

[0776] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0777] Step 2:

[0778] The server receives the PDF file and temporarily saves it to storage. For example, the save location is set to / tmp / uploaded_paper.pdf, and the file path is checked when passing it to the next process.

[0779] Step 3:

[0780] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Text is extracted from each page and combined into a single continuous text file. Images are saved to a specific directory, and mathematical formulas are output as images.

[0781] Step 4:

[0782] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. For example, text can be sent using an API, and the results can be saved in text format.

[0783] python

[0784] import openai

[0785] openai.api_key = 'your-api-key'

[0786] response = openai.Completion.create(

[0787] engine="text-davinci-003",

[0788] prompt="Analyze and summarize the following text:\n" + all_text,

[0789] max_tokens=500

[0790] )

[0791] summary = response.choices[0].text

[0792] Preservation process

[0793] Step 5:

[0794] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts graph elements and features from each image and saves information such as model parameters and activation functions in text format.

[0795] python

[0796] from yolov5 import YOLOv5

[0797] yolo = YOLOv5("path / to / yolov5 / weights")

[0798] for img_path in image_save_paths:

[0799] results = yolo.predict(img_path)

[0800] parsed_results = results.pandas().xyxy Save process

[0801] Step 6:

[0802] The server extracts mathematical formulas and analyzes them using a mathematical formula analysis tool (e.g., Mathpix). OCR technology is used to convert the mathematical formula images into text format, and the resulting text data is saved in a format that can be understood.

[0803] python

[0804] import pytesseract

[0805] Import Image from PIL

[0806] math_images = [] Assumes that the images of mathematical formulas have been identified and saved in advance.

[0807] math_texts = [pytesseract.image_to_string(Image.open(img_path), config="math_configuration") for img_path in math_images]

[0808] Interpretation and storage

[0809] Step 7:

[0810] The server uses an emotion engine to recognize the user's emotions in real time. Emotional data is acquired based, for example, on user input, interactions, and video analysis from the webcam. Once emotional data is acquired, the analysis method and the way results are provided are adjusted accordingly.

[0811] python

[0812] import emotion_recognition

[0813] emotions = emotion_recognition.recognize_webcam()

[0814] Preservation process

[0815] Step 8:

[0816] The server integrates the analysis results of text, images, and mathematical formulas to generate the program code necessary for reproduction experiments. It combines the results of natural language processing models, image analysis, and mathematical formula analysis to generate Python code that enables reproduction experiments. In addition, it adds guides and explanations tailored to the user's emotions.

[0817] Step 9:

[0818] The server provides the user with the generated analysis results and program code. The method of delivery is adjusted according to the user's preferences; for example, a detailed step-by-step tutorial may be provided. The user then uses this to perform reproduction experiments.

[0819] (Example 2)

[0820] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0821] In modern research, it is crucial to quickly and accurately understand the content of research papers and to conduct replication experiments. However, manually analyzing a vast number of papers and extracting the information necessary for summarization and replication experiments is an extremely time-consuming task. Furthermore, providing information that disregards the user's feelings is ineffective in understanding and utilizing the analysis results. This makes it difficult for researchers to conduct replication experiments efficiently, leading to a slowdown in research progress.

[0822] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0823] In this invention, the server includes means for receiving electronic documents from a user, means for extracting document information, image information, and mathematical formula information from the electronic documents, means for using a natural language processing model to analyze the extracted document information, means for using an image analysis model to analyze the extracted image information, means for using a mathematical formula analysis tool to analyze the extracted mathematical formula information, means for integrating the analysis results and generating program code necessary for experimental reproduction, means for recognizing the user's emotions and adjusting the method and order of presentation of the analysis results, and means for providing the user with the analysis results and the generated program code. This makes it possible to quickly and accurately analyze the content of a paper and efficiently provide the information necessary for reproduction experiments. By providing information in accordance with the user's emotions, it is expected that the understanding of the analysis results will be enhanced and the reproduction experiments will proceed smoothly.

[0824] A "user" is an individual or organization that uses the system to upload electronic documents and receives analysis results and program code.

[0825] An "electronic document" refers to documents such as academic papers and research materials that are saved in PDF format or other digital formats.

[0826] "Document information" refers to text data extracted from electronic documents.

[0827] "Image information" refers to image data extracted from electronic documents.

[0828] "Mathematical formula information" refers to mathematical formula data extracted from electronic documents.

[0829] A "natural language processing model" is a machine learning model or algorithm used to analyze extracted document information and generate summaries and key points.

[0830] An "image analysis model" is a machine learning model or algorithm used to analyze extracted image information and extract the constituent elements and features of a figure or chart.

[0831] A "mathematical formula analysis tool" is software or algorithms that analyze extracted mathematical formula information and generate the meaning and application examples of the formulas in text format.

[0832] "Analysis results" refer to the results obtained using natural language processing models, image analysis models, and mathematical analysis tools.

[0833] "Program code" refers to the source code generated based on the analysis results, used to execute the reproduction experiment.

[0834] "Emotion recognition" refers to the process and technology of analyzing a user's emotions and adjusting the way results are presented and the order in which they are presented.

[0835] This invention is a system that analyzes electronic documents and comprehensively understands document information, image information, and mathematical formula information. This system allows users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system primarily operates on the server side, analyzing electronic documents uploaded by users and providing the results.

[0836] The server performs the following process: First, the user uploads an electronic document via the web interface. Then, the server receives this electronic document and temporarily stores it in storage. A possible specific storage location might be / tmp / uploaded_document.pdf.

[0837] Next, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information. Each page is analyzed separately, and the text is saved as a single continuous document, while images and mathematical formulas are saved in separate directories (e.g., / tmp / images / ).

[0838] The extracted document information is analyzed using a natural language processing model (e.g., GPT-4). This analysis generates a summary and key points of the paper, and the results are saved in text format (e.g., / tmp / summary.txt). Similarly, the extracted image information is analyzed using an image analysis model (e.g., YOLO, CNN), and the information obtained from the images is saved in text format (e.g., / tmp / image_analysis.txt). In addition, mathematical formula information is analyzed using a mathematical formula analysis tool (e.g., Mathpix), and the meaning and application examples of the formulas are described in natural language and saved as text data (e.g., / tmp / formula_analysis.txt).

[0839] Furthermore, the server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed in real time through analysis of the user's past interactions and video from the webcam (if permitted). It determines whether the user is showing anxiety or interest and stores the results in an internal data structure.

[0840] These analysis results are integrated to generate the program code necessary for reproduction experiments. The generated code includes the model definition, data preprocessing, and training procedures. For example, the generated code is saved in / tmp / reproduction_code.py. During this process, the server adjusts the order in which results are presented and the level of detail in the explanations based on the user's emotions. For example, if the user is feeling anxious, it provides detailed explanations and step-by-step guidance.

[0841] Finally, the server provides the user with the generated analysis results and program code. The user can download these from the web interface and use them as reference information for conducting reproduction experiments.

[0842] As a concrete example of operation, the following prompt statements are possible:

[0843] "I have uploaded a PDF of a paper on a new gradient descent method. Please analyze the contents of this PDF. Generate a summary of the text, images, and mathematical formulas, and then create Python code for a replication experiment based on these. Also, please adjust the presentation method of the analysis results based on my interests and concerns."

[0844] As described above, this system helps advance research activities by reducing the burden on users and efficiently providing analysis results.

[0845] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0846] Step 1:

[0847] Users upload electronic documents (PDF files) to the server via a web interface.

[0848] Specifically, the user opens a browser and clicks the file upload button. They select a PDF file from the file selection dialog and click the "Upload" button.

[0849] Input: PDF file uploaded by the user

[0850] Output: PDF file sent to the server

[0851] Step 2:

[0852] The server receives the uploaded PDF file and temporarily stores it in storage.

[0853] Specifically, the file is saved to / tmp / uploaded_document.pdf.

[0854] Input: PDF file sent by the user to the server

[0855] Output: PDF file saved in the specified directory on the server.

[0856] Step 3:

[0857] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information.

[0858] Specifically, each page is analyzed, the text is saved as a continuous document, and images and mathematical formulas are saved in a separate directory (e.g., / tmp / images / ).

[0859] Input: Saved PDF file

[0860] Output: A dataset of extracted text, images, and mathematical formulas.

[0861] Step 4:

[0862] The server uses a natural language processing model (e.g., GPT-4) to analyze the extracted document information.

[0863] Specifically, the extracted text is input into a text analysis model, which then generates a summary and key points. The generated analysis results are saved in text format to / tmp / summary.txt.

[0864] Input: Extracted text data

[0865] Output: A text file containing summaries and key points.

[0866] Step 5:

[0867] The server analyzes image information using an image analysis model (e.g., YOLO, CNN).

[0868] Specifically, the image analysis model is run using each image file as input, and the information obtained from the analysis is saved to / tmp / image_analysis.txt.

[0869] Input: Extracted image data

[0870] Output: Text file containing image analysis results

[0871] Step 6:

[0872] The server uses a mathematical analysis tool (e.g., Mathpix) to analyze the mathematical information.

[0873] Specifically, the tool takes an image of a mathematical formula as input and converts it into text format. It then describes the meaning and application examples of the formula in natural language and saves them to / tmp / formula_analysis.txt.

[0874] Input: Extracted formula data

[0875] Output: A text file containing the results of the mathematical formula analysis.

[0876] Step 7:

[0877] The server uses an emotion engine to recognize the user's emotions.

[0878] Specifically, the system analyzes user input, interaction data, and, if permitted, webcam footage to determine whether the user is showing anxiety or interest. The results are stored in an internal data structure.

[0879] Input: User interaction data and video data

[0880] Output: Data on the user's emotional state

[0881] Step 8:

[0882] The server integrates the analysis results so far and generates the program code necessary for the reproduction experiment.

[0883] Specifically, the system combines the results of text analysis, image analysis, and mathematical analysis into a single file. It then incorporates a summary of the natural language processing model to generate Python code that allows for reproduction experiments. The generated code is saved in ` / tmp / reproduction_code.py`. Furthermore, the system adjusts the presentation order and level of detail of the results based on the user's emotions.

[0884] Input: Text analysis results, image analysis results, mathematical formula analysis results, user emotional state data

[0885] Output: Python code for reproduction experiment

[0886] Step 9:

[0887] The server provides the user with the generated analysis results and program code.

[0888] Specifically, the analysis results and code are presented as download links on a web interface. Users download them and perform reproduction experiments.

[0889] Input: Generated analysis results and program code

[0890] Output: Analysis results and program code provided to the user.

[0891] (Application Example 2)

[0892] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server," and the smart glasses 214 will be referred to as the "terminal."

[0893] Currently, it is difficult for workers in logistics centers to quickly understand information about new technologies and algorithms and perform tasks efficiently based on that understanding. Furthermore, because there is no system in place to understand and appropriately respond to workers' emotions, anxiety and stress increase when the learning curve is steep, leading to decreased work efficiency. Therefore, there is a need to develop a system that can quickly and accurately analyze the content of research papers and adjust the method of information delivery according to the workers' emotions.

[0894] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for receiving a PDF of a paper from a user, means for extracting text, images, and mathematical formulas from the PDF, means for using a natural language processing model to analyze the extracted text, means for using an image analysis model to analyze the extracted images, means for using a mathematical formula analysis tool to analyze the extracted mathematical formulas, means for recognizing the user's emotions, means for integrating the analysis results and generating program code necessary for reproduction experiments, means for adjusting the method and presentation order of the analysis results based on the user's emotions, and means for providing the user with the analysis results and the generated program code. This enables an assist system that allows workers in a logistics center to quickly understand new technical information and perform work efficiently.

[0895] A "user" is a person or entity that uses the system to upload a PDF of a research paper and receive the analysis results.

[0896] A "paper" is a document that describes the results of academic research or studies in PDF format.

[0897] "PDF" stands for Portable Document Format, a file format that can contain text, images, and mathematical formulas in a document.

[0898] "Text" refers to the written information contained in a PDF, including sentences, explanations, and comments.

[0899] An "image" is an element that represents visual information such as graphs, diagrams, and photographs included within a PDF document.

[0900] "Mathematical formulas" refer to mathematical expressions written within a PDF document, including calculation formulas and equations.

[0901] A "natural language processing model" is an artificial intelligence technology used to analyze, summarize, and interpret text, such as a model that understands the meaning of extracted text.

[0902] An "image analysis model" is an artificial intelligence technology used to analyze images; for example, it is a model used to extract and understand information contained in an image.

[0903] A "mathematical formula analysis tool" is a tool used to analyze mathematical formulas and understand their meaning and application examples.

[0904] "Emotion recognition" is a technology that analyzes and identifies a user's emotions, and is about understanding the user's state of mind.

[0905] "Analysis results" refer to information and content generated based on the analysis of text, images, and mathematical formulas extracted from the PDF.

[0906] "Program code" refers to computer language code that describes the procedures and operations necessary to carry out a reproduction experiment.

[0907] "Means for adjusting the method of provision and presentation order" refers to a function that changes the way analysis results are displayed and the order in which they are presented based on the user's emotions.

[0908] To implement this invention, it is necessary to build a system that includes the following steps. Specific hardware includes a head-mounted display (HMD) for user use, a server, and a webcam. Software used may include PDF parsing libraries (e.g., PyMuPDF or PDFMiner), OCR software (e.g., Pytesseract), natural language processing models (e.g., Hugging Face Transformers), emotion recognition engines (e.g., EmotionRecognizer), and image analysis models (e.g., YOLO or CNN).

[0909] First, the user uploads the PDF of their paper to the server via the HMD (Head-Mounted Display). The server receives the PDF file and temporarily stores it in storage. This file is later broken down into text, images, and mathematical formulas using a PDF analysis library. The text is extracted using OCR software, and the images are saved in a separate directory. The mathematical formulas are saved as images and analyzed using a mathematical formula analysis tool.

[0910] Next, the server uses a natural language processing model to analyze the extracted text and generate a summary and key points of the paper. An image analysis model is used to extract elements and features of figures and tables from images, and the resulting information is saved in text format. A mathematical formula analysis tool is used to obtain text information from mathematical formula images, explaining the meaning and application examples of the formulas.

[0911] The server also uses a webcam to recognize the user's emotions in real time. The emotion recognition engine determines emotions from the user's input, interactions, and facial expressions, and evaluates whether the user is showing anxiety or interest.

[0912] The server integrates these analysis results and automatically generates the Python code necessary for replication experiments. It adjusts the presentation order and detailed explanations of the analysis results based on the user's emotional state. For example, if the user is feeling anxious, it provides a detailed step-by-step guide to help them confidently replicate the experiment.

[0913] Finally, the generated analysis results and program code are provided to the user via the HMD. The user can then use this as a reference to perform reproduction experiments.

[0914] As a concrete example, when a worker at a logistics center uploads a paper on a new cargo placement algorithm, the analysis results are displayed on a head-mounted display. If the worker has any questions or concerns, an emotion recognition engine detects this and automatically provides a detailed step-by-step guide.

[0915] Examples of prompts include: "Analyze the key points of the latest research paper on logistics centers. A detailed explanation of specific cargo placement algorithms is required. Also, suggest ways to alleviate worker anxiety."

[0916] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0917] Step 1:

[0918] The user uploads a PDF of their paper to the server via a head-mounted display. The input is the PDF file selected by the user, and the output is this PDF file being saved on the server. The server temporarily stores the file in storage and passes it on to the next process.

[0919] Step 2:

[0920] The server parses a PDF file and extracts text, images, and mathematical formulas. The input is a PDF file stored on the server, and the output is the extracted text, images, and mathematical formulas. Specifically, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to process each page, concatenating the text into a single continuous text file, saving the images in a separate directory, and saving the mathematical formulas as images.

[0921] Step 3:

[0922] The server uses a natural language processing model to analyze the extracted text. The input is the extracted text, and the output is a summarized text or key points. Specifically, the server uses a natural language processing model (e.g., Hugging Face Transformers) to summarize the main sections of a paper and extract the core of a new method.

[0923] Step 4:

[0924] The server uses an image analysis model to analyze the extracted images. The input is the extracted images, and the output is text data of the constituent elements and features derived from the images. Specifically, the server uses an image analysis model (e.g., YOLO or CNN) to analyze the components of graphs and charts from the images and extract parameters and activation functions for each layer.

[0925] Step 5:

[0926] The server uses a mathematical expression analysis tool to analyze the extracted mathematical formulas. The input is an image of the formula, and the output is an interpretation of the formula in text format. Specifically, the server uses a mathematical expression analysis tool (e.g., Mathpix) to convert the image of the formula into text format and explains the meaning and application examples of the formula.

[0927] Step 6:

[0928] The server uses a webcam to recognize the user's emotions. The input is the user's video captured by the webcam, and the output is the estimated emotion of the user. Specifically, the server uses an emotion recognition engine (e.g., EmotionRecognizer) to determine the emotion in real time from the user's facial expressions and interactions.

[0929] Step 7:

[0930] The server integrates the analysis results to date and generates the program code necessary for the replication experiment. Inputs include analysis results of text, images, and mathematical formulas, as well as user sentiment information. Outputs are Python code required for the replication experiment. Specifically, the server uses a generative AI model to integrate the results and automatically generates program code that enables the replication experiment.

[0931] Step 8:

[0932] The server adjusts the method and order in which analysis results are presented based on the user's emotions. Input is the user's emotion information and generated program code, while output is the analysis results and program code formatted according to the emotion. Specifically, the server adds detailed explanations and step-by-step guides based on the emotion recognition results.

[0933] Step 9:

[0934] The server provides the user with the generated analysis results and program code. The input is the adjusted analysis results and program code, and the output is the result displayed on the user's head-mounted display. Specifically, the server allows the user to download or view the analysis results and code via a web interface.

[0935] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[0936] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0937] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the smart glasses 214.

[0938] [Third Embodiment]

[0939] Figure 5 shows an example of the configuration of the data processing system 310 according to the third embodiment.

[0940] As shown in Figure 5, the data processing system 310 includes a data processing device 12 and a headset terminal 314. An example of the data processing device 12 is a server.

[0941] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0942] The headset terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication interface 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, camera 42, and display 343 are also connected to the bus 52.

[0943] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[0944] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[0945] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[0946] Figure 6 shows an example of the main functions of the data processing device 12 and the headset terminal 314. As shown in Figure 6, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[0947] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0948] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0949] In the headset terminal 314, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0950] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the headset terminal 314 will be referred to as the "terminal".

[0951] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. This system operates on a server side and analyzes PDFs uploaded by users, providing the results.

[0952] Program processing flow

[0953] 1. Receiving the PDF

[0954] Users upload their research papers as PDFs to the server via a web interface. This serves as the entry point to the system.

[0955] 2. Data Extraction

[0956] The server saves the received PDF file to a specific directory. Next, a PDF parsing library is used to extract text, images, and mathematical formulas from the PDF.

[0957] The text is read page by page, combined, and saved as a single continuous text file. Images are saved in a specific folder, and mathematical formulas are converted into text format using OCR technology.

[0958] 3. Text Analysis

[0959] The server uses a natural language processing model to analyze the extracted text. This generates a summary and key points of the paper.

[0960] 4. Image Analysis

[0961] The server uses an image analysis model to analyze images within the paper. It retrieves data points from graphs and figures contained within the images, as well as their interpretations, in text format.

[0962] 5. Formula Analysis

[0963] The server uses a mathematical analysis tool to analyze the extracted mathematical formulas. The analysis results are saved as text data, including interpretations of the formulas' meanings and applications.

[0964] 6. Support for Integration and Reproducibility Experiments

[0965] The server integrates the analysis results and generates code for replication experiments. This allows users to obtain specific procedures and program code to reproduce the experiment based on the content of the paper.

[0966] 7. Providing the results

[0967] The server returns the generated analysis results and program code to the user. The user can then use this to gain a deeper understanding of the paper's content and smoothly carry out replication experiments.

[0968] Specific example

[0969] As a concrete example, we can consider the following scenario.

[0970] 1. Receiving the PDF

[0971] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[0972] 2. Data Extraction

[0973] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[0974] 3. Text Analysis

[0975] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[0976] 4. Image Analysis

[0977] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[0978] 5. Formula Analysis

[0979] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[0980] 6. Support for Integration and Reproducibility Experiments

[0981] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[0982] 7. Providing the results

[0983] The server provides the user with a generated summary, image analysis results, mathematical formula interpretations, and code for replication experiments. The user can then use this information to reproduce the paper's experiments in their own environment.

[0984] As described above, the system of the present invention enables users to deeply understand the content of research papers and efficiently conduct reproduction experiments.

[0985] The following describes the processing flow.

[0986] Step 1:

[0987] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[0988] Step 2:

[0989] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[0990] Step 3:

[0991] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[0992] Step 4:

[0993] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. The analysis results are saved in text format and used for subsequent processing.

[0994] Step 5:

[0995] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts the constituent elements and features of each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[0996] Step 6:

[0997] The server extracts mathematical formulas and analyzes them using a mathematical analysis tool (e.g., Mathpix). The formula images are converted into text format, and the meaning and application examples of the formulas are described in natural language. This allows the content of the formulas to be saved as text data.

[0998] Step 7:

[0999] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[1000] Step 8:

[1001] The server provides the user with the generated analysis results and program code. The user downloads this from the web interface and uses it as a reference to conduct reproduction experiments.

[1002] (Example 1)

[1003] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[1004] Conventional technologies required advanced expertise and considerable time to comprehensively analyze the content of papers and documents and facilitate reproduction experiments. Furthermore, the process of integrating results using multiple analysis methods and tools was complex and burdensome for users. Therefore, the present invention aims to provide a system that automates document analysis and preparation for reproduction experiments, enabling efficient and accurate execution.

[1005] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[1006] In this invention, the server includes means for receiving a document file from a user, means for extracting character data, figures, and formulas from the file, means for using a natural language processing model to analyze the extracted character data, means for using an image analysis model to analyze the extracted figures, means for using an analysis tool to analyze the extracted formulas, means for integrating the analysis results and generating a program necessary for reproduction verification, and means for providing the user with the analysis results and the generated program. This makes it possible to comprehensively analyze the contents of a document and to perform reproduction experiments efficiently and accurately.

[1007] A "user" refers to an entity that uploads document files to the server and receives the analysis results or programs.

[1008] "Document" refers to electronic files containing text information such as papers and reports, including formats such as PDF.

[1009] A "file" refers to digital data in which the contents of a document are stored.

[1010] "Character data" refers to the text information contained within a document.

[1011] "Graphics" refers to visual information such as images, graphs, and charts included within a document.

[1012] "Formula" refers to mathematical formulas or mathematical expressions included in a document.

[1013] "Extraction" refers to the process of taking specific data (text data, graphics, formulas) out of a file.

[1014] A "natural language processing model" refers to a machine learning model or algorithm used to analyze extracted text data and extract important information.

[1015] An "image analysis model" refers to a machine learning model or algorithm that analyzes extracted graphic data and converts visual information into text format.

[1016] An "analysis tool" refers to software or algorithms used to analyze extracted formulas and understand their meaning.

[1017] "Integration" refers to the process of combining analysis results into a single format and ensuring consistency.

[1018] "Reproduction verification" refers to the process of reproducing experiments or simulations based on the content of a document.

[1019] A "program" refers to a set of code and procedures necessary for reproducing and verifying a problem.

[1020] "Results" refers to the information obtained from the analysis of the document.

[1021] This invention is a document analysis and reproduction experiment support system, primarily configured to operate on the server side. The specific implementation methods for each step are described below.

[1022] Receiving document files

[1023] Users upload document files to the server using a web interface. These document files include PDFs and other formats containing text information such as papers and reports. The server saves the received document files to a specific directory (e.g., / tmp / pdf_uploads / ).

[1024] Data extraction

[1025] The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text data, shapes, and formulas from the file. The specific steps are as follows:

[1026] Text data: Extracts text from each page of a document and saves it as a series of text data.

[1027] Shapes: Extracts images, graphs, and charts from a document and saves them, for example, to a folder such as / tmp / pdf_uploads / images / .

[1028] Formula: Convert mathematical formulas to text format using an OCR library (e.g., Mathpix) and save them.

[1029] Data analysis

[1030] Analysis of character data

[1031] The server uses natural language processing models (e.g., BERT, GPT) to analyze text data and extract summaries and key points from papers. This allows the server to generate summaries that enable users to quickly understand the content of documents.

[1032] Analysis of geometric shapes

[1033] The server analyzes the shapes using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format. The analysis results include, for example, interpretations of the model's architecture diagram and experimental results graphs.

[1034] Analysis of the formula

[1035] The server uses a mathematical analysis tool (e.g., SymPy) to analyze mathematical formulas and generate interpretations of their meaning and applications. The analysis results are saved as text data explaining the meaning of the formulas.

[1036] Support for integration and replication experiments

[1037] The server integrates these analysis results and generates a program for replication experiments. This program is provided as Python code that includes model definition, data preprocessing, and training procedures. Users can use this program to replicate the experiment in their own environment.

[1038] Providing results

[1039] The server returns the generated analysis results and program code to the user. The analysis results include a summary, image analysis results, interpretation of mathematical formulas, and code for reproduction experiments, enabling the user to gain a deeper understanding of the document's content and efficiently conduct reproduction experiments.

[1040] Specific example

[1041] As a concrete example, consider a case where a user uploads a paper on a new learning method for AI models. The server can perform the analysis through the following steps:

[1042] The server extracts 40 pages of text, 5 images, and 3 mathematical formulas, and analyzes each of them.

[1043] The server uses a natural language processing model to generate a summary that reads, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[1044] The server uses an image analysis model to analyze the image, concluding that "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer."

[1045] The server uses a mathematical analysis tool to interpret Equation 1 as "showing the loss function of the proposed method, which minimizes the squared error."

[1046] The server generates Python code for reproduction experiments and provides it to the user.

[1047] Examples of prompt statements

[1048] "Analyze a paper on a new AI model learning method, extract key points, interpret image content and mathematical formulas, and generate code for replication experiments."

[1049] "Please analyze the following PDF paper and provide the summary, data points for the graphs, meaning of the formulas, and experimental reproduction procedures. PDF link: [link]"

[1050] As described above, the document analysis system of the present invention can provide an effective means for users to deeply understand the content of documents and efficiently conduct reproduction experiments.

[1051] The flow of the specific processing in Example 1 will be explained using Figure 11.

[1052] Step 1:

[1053] PDF reception

[1054] Input: The PDF file to be uploaded by the user.

[1055] Specific operation: On the web interface, the user clicks the "Select File" button, selects the desired PDF file from their local disk, and clicks the upload button.

[1056] Data processing: The server saves the received PDF files to a specific directory (e.g., / tmp / pdf_uploads / ).

[1057] Output: PDF file saved in / tmp / pdf_uploads / .

[1058] Step 2:

[1059] Data extraction

[1060] Input: Saved PDF file.

[1061] Specific operation: The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text, images, and mathematical formulas from PDF files.

[1062] Text extraction: Reads text from each page and saves it as a series of text files (e.g., / tmp / pdf_uploads / extracted_text.txt).

[1063] Image extraction: Identifies images within the PDF and saves them to a specific folder (e.g., / tmp / pdf_uploads / images / ).

[1064] Formula extraction: Use an OCR library (e.g., Mathpix) to convert formulas in the PDF into text format and save it (e.g., / tmp / pdf_uploads / extracted_equations.txt).

[1065] Data processing: Extraction of text, images, and mathematical formulas using a PDF analysis library.

[1066] Output: Text data ( / tmp / pdf_uploads / extracted_text.txt), image data ( / tmp / pdf_uploads / images / ), mathematical formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[1067] Step 3:

[1068] Text analysis

[1069] Input: Extracted text data ( / tmp / pdf_uploads / extracted_text.txt).

[1070] Specific operation: The server uses a natural language processing model (e.g., BERT, GPT) to analyze the text data and extract summaries and key points.

[1071] Section Identification: Analyzes text data to identify sections of a paper (e.g., Introduction, Methods, Results, Discussion, etc.).

[1072] Summary generation: Generate summaries for each section and extract the core elements of the new learning method.

[1073] Data processing: Text analysis and summarization using natural language processing models.

[1074] Output: Summarized text data.

[1075] Step 4:

[1076] Image analysis

[1077] Input: Extracted image data ( / tmp / pdf_uploads / images / ).

[1078] Specific operation: The server analyzes images using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format.

[1079] Graph and chart analysis: Extracting data points and interpretations from images (e.g., "Figure 1 shows a 3-layer convolutional neural network, with ReLU used in each layer").

[1080] Data processing: Analysis of image data using image analysis models.

[1081] Output: Text data of the image analysis results.

[1082] Step 5:

[1083] Formula analysis

[1084] Input: Extracted formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[1085] Specific operation: The server uses a mathematical expression analysis tool (e.g., SymPy) to analyze the formula and generate an interpretation of its meaning and application.

[1086] Mathematical analysis: Analyze the mathematical formula and interpret it as "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[1087] Data calculation: Analysis of mathematical formulas using mathematical analysis tools.

[1088] Output: Text data of the mathematical analysis results.

[1089] Step 6:

[1090] Support for integration and replication experiments

[1091] Input: Text analysis results, image analysis results, mathematical formula analysis results.

[1092] Specific operation: The server integrates these analysis results and generates a program for reproduction experiments.

[1093] Integration of analysis results: Organize the analysis results and compile them into a single document (e.g., Markdown format).

[1094] Program generation: Generates Python code for replication experiments, describing the model definition, data preprocessing, and training procedures.

[1095] Data processing: Integration of analysis results and generation of program code.

[1096] Output: Integrated analysis results document and Python code.

[1097] Step 7:

[1098] Providing results

[1099] Input: Integrated analysis results document and Python code.

[1100] Specific operation: The server compresses the generated analysis results and program code into a zip file and provides a download link for the user.

[1101] Output: A download link that the user can access.

[1102] The above describes the specific operation and input / output of each processing step in this system.

[1103] (Application Example 1)

[1104] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[1105] When new production technologies or methods are proposed within a factory, actually reproducing and verifying them requires time and effort. Furthermore, understanding technical documents such as research papers requires specialized knowledge, and appropriately interpreting their content and reflecting it in experiments is difficult. As a result, the introduction of new technologies is often delayed, hindering improvements in production efficiency and quality.

[1106] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[1107] In this invention, the server includes means for receiving a PDF of a technical document from a user; means for extracting information, images, and mathematical representations from the PDF; means for using a natural language processing model to analyze the extracted information; means for using an image processing model to analyze the extracted images; means for using a mathematical expression analysis tool to analyze the extracted mathematical representations; means for integrating the analysis results and generating program code and automated experimental procedures necessary for reproduction experiments; and means for providing the analysis results and generated program code to the user. This makes it possible to quickly analyze new technologies described in technical documents within a factory and to automatically perform reproduction experiments.

[1108] A "user" is an individual or organization that operates the system to analyze technical documents or conduct reproduction experiments.

[1109] "Technical documents" refer to documents containing content related to science, technology, or engineering, such as research papers and technical reports.

[1110] "PDF" is an abbreviation for Portable Document Format, a file format for electronically displaying the contents of a document.

[1111] "Information" refers collectively to the text, images, and mathematical formulas included in technical documents.

[1112] "Images" refer to visual data such as diagrams, graphs, and photographs included within technical documents.

[1113] "Mathematical expression" refers to mathematical formulas and statistical descriptions included in technical documents.

[1114] A "natural language processing model" is a computational model used to analyze text data and understand its meaning and structure.

[1115] An "image processing model" is a computational model used to analyze image data and extract its content and features.

[1116] A "mathematical formula analysis tool" is software used to analyze mathematical formulas and understand their meaning and structure.

[1117] "Analysis results" refers to the collective data obtained by natural language processing models, image processing models, and mathematical analysis tools.

[1118] "Program code" refers to a computer program that describes the procedures and processes necessary to carry out a reproduction experiment.

[1119] An "automated experimental procedure" is a specific set of steps generated based on analysis results, which are used to perform a reproducible experiment using automated equipment.

[1120] A "system" is a general term for a complex set of devices or software that automates everything from the analysis of technical documents to reproduction experiments.

[1121] This invention provides a system that analyzes PDF technical documents and comprehensively analyzes text, images, and mathematical representations to facilitate users in performing replication experiments. This system includes a paper decoding system, a server, a cloud server, and a factory robot.

[1122] System Overview

[1123] Program processing

[1124] The server first receives technical document PDFs from users. The received PDFs are saved in a specific directory. Next, a PDF parsing library (e.g., PyPDF2) is used to extract text, images, and mathematical representations from the PDFs. The extracted text is analyzed using a natural language processing model (e.g., spacy) to generate summaries and key points of the papers. Images are analyzed using an image processing model (e.g., pytesseract) to convert the included data and graphs into text format. Mathematical representations are analyzed using a mathematical expression analysis tool (e.g., regular expressions) to explain their meaning.

[1125] The analysis results are integrated to generate the program code and automated experimental procedures necessary for reproduction experiments. This generated code is provided to the user. Furthermore, the system provides procedures and tools for automatically performing reproduction experiments based on the analysis results.

[1126] Hardware and software used

[1127] Hardware:

[1128] Factory robots (e.g., KUKA robots, ABB robots)

[1129] Cloud servers (AWS, Google Cloud, etc.)

[1130] software:

[1131] PDF analysis: PyPDF2

[1132] Natural Language Processing: Spacy

[1133] Image analysis: pytesseract

[1134] Formula extraction: Regular expressions (re)

[1135] Specific example

[1136] As a concrete example, consider a scenario in which a technical document on a new, highly efficient production technology is introduced to a factory. First, the factory manager uploads the PDF of the technical document to a factory robot using a smartphone. Analysis is performed on a cloud server, and the following results are generated.

[1137] Summary: "This technical document proposes a new technology that achieves a 20% improvement in efficiency compared to conventional technology."

[1138] Image analysis: "Figure 1 shows a flowchart of the new technology, and the main steps are as follows..."

[1139] Mathematical analysis: "Equation 1 shows the optimization parameters for improving efficiency."

[1140] Based on these analysis results, detailed program code necessary for reproduction experiments is generated. This generated code is provided to the factory robot, which automatically executes the reproduction experiment. This process allows for rapid verification of the effectiveness of new technologies and supports their implementation within the factory.

[1141] Example of a prompt

[1142] Examples of specific prompt messages are as follows:

[1143] "Please write a program for industrial robots to analyze papers on new production technologies and automate the replication experiments."

[1144] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[1145] Step 1:

[1146] PDF reception

[1147] The user uploads a PDF file of a technical document. This PDF is received by the server and saved to a specific directory.

[1148] Input: PDF file of a technical document.

[1149] Output: PDF file saved in the server directory.

[1150] Specific operation: When a user selects a PDF from the web interface and presses the upload button, the server saves the file to the specified directory.

[1151] Step 2:

[1152] Data extraction

[1153] The server uses the PDF parsing library (PyPDF2) to extract text, images, and mathematical representations from PDF files.

[1154] Input: Saved PDF file.

[1155] Output: Extracted text files, a list of image files, and a list of mathematical representations.

[1156] Specific operation: The server opens the PDF file, extracts text from each page, saves the image files, and adds the mathematical representations to a list.

[1157] Step 3:

[1158] Text analysis

[1159] The server uses a natural language processing model (SPacy) to analyze the extracted text.

[1160] Input: Extracted text file.

[1161] Output: Text summary and key points.

[1162] Specific operation: The server inputs a text file into a natural language processing model and extracts a summary and important sections.

[1163] Step 4:

[1164] Image analysis

[1165] The server uses an image processing model (pytesseract) to analyze the extracted images.

[1166] Input: List of extracted image files.

[1167] Output: Text data extracted from the image.

[1168] Specific operation: The server sequentially inputs each image file into the image processing model and extracts the text data.

[1169] Step 5:

[1170] Formula analysis

[1171] The server uses a mathematical expression analysis tool (regular expressions) to analyze the extracted mathematical representation.

[1172] Input: A list of mathematical expressions.

[1173] Output: Interpretation and explanation of the meaning of the formula.

[1174] Specific operation: The server inputs the list of mathematical formulas into a regular expression tool and generates text explaining the meaning of each formula.

[1175] Step 6:

[1176] Support for integration and replication experiments

[1177] The server integrates the analysis results and generates the program code and automated experimental procedures necessary for reproduction experiments.

[1178] Input: Text summaries, text data extracted from images, and interpretation of mathematical formulas.

[1179] Output: Integrated analysis results and program code for replication experiments.

[1180] Specific operation: The server integrates all data and generates Python code for reproduction experiments based on the analysis results.

[1181] Step 7:

[1182] Providing results

[1183] The server provides the user with the generated analysis results and program code.

[1184] Input: Integrated analysis results and program code for replication experiments.

[1185] Output: Analysis results and program code provided to the user.

[1186] Specific operation: The server formats the analysis results and code and provides them to the user as a download link. It also sends the analysis results via email or other methods as needed.

[1187] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[1188] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system operates on a server side, analyzing PDFs uploaded by users and providing the results.

[1189] Program processing flow

[1190] 1. Receiving the PDF

[1191] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[1192] 2. Data Extraction

[1193] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[1194] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[1195] 3. Text Analysis

[1196] The server utilizes natural language processing models (e.g., GPT-4) to analyze the extracted text. This generates summaries and key points of the papers. The analysis results are saved in text format and used for subsequent processing.

[1197] 4. Image Analysis

[1198] The server performs the analysis using an image analysis model (e.g., YOLO, CNN). It extracts the components and features of the figures and tables from each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[1199] 5. Formula Analysis

[1200] The server uses a mathematical formula analysis tool (e.g., Mathpix) to perform the analysis. It converts the mathematical formula image into text format and describes the meaning and application examples of the formula in natural language. This allows the content of the formula to be saved as text data.

[1201] 6. Emotion recognition

[1202] The server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed through user input, interaction, and video analysis from the webcam. For example, it can determine in real time whether the user is showing anxiety or interest.

[1203] 7. Support for Integration and Reproducibility Experiments

[1204] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[1205] Furthermore, the method of providing analysis results will be adjusted based on the user's emotions. For example, if a user is feeling anxious, more detailed explanations and step-by-step guides will be provided.

[1206] 8. Providing results

[1207] The server provides the user with the generated analysis results and program code. The order and method of presenting the results are adjusted according to the user's mood. For example, if the user appears interested, important results can be presented first to maintain their attention.

[1208] Users download this from the web interface and use it as a reference for conducting reproduction experiments. The reproduction experiment process proceeds based on the provided code and explanations.

[1209] Specific example

[1210] As a concrete example, we can consider the following scenario.

[1211] 1. Receiving the PDF

[1212] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[1213] 2. Data Extraction

[1214] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[1215] 3. Text Analysis

[1216] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[1217] 4. Image Analysis

[1218] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[1219] 5. Formula Analysis

[1220] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[1221] 6. Emotion recognition

[1222] The server recognizes the user's emotions and determines whether the user is feeling anxious about the analysis results. For example, if a user has asked many questions about past analysis results, the server will determine that the user is feeling anxious.

[1223] 7. Support for Integration and Reproducibility Experiments

[1224] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[1225] At the same time, the order in which the analysis results are presented and the amount of detailed explanation are adjusted based on the user's emotions.

[1226] 8. Providing results

[1227] The server provides the user with generated summaries, image analysis results, mathematical formula interpretations, and code for reproduction experiments.

[1228] Users receive information adjusted through emotion recognition and can utilize the analysis results in a format that suits their own emotions, thereby facilitating the smooth progress of replication experiments.

[1229] The following describes the processing flow.

[1230] Step 1:

[1231] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[1232] Step 2:

[1233] The server receives the PDF file and temporarily saves it to storage. For example, the save location is set to / tmp / uploaded_paper.pdf, and the file path is checked when passing it to the next process.

[1234] Step 3:

[1235] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Text is extracted from each page and combined into a single continuous text file. Images are saved to a specific directory, and mathematical formulas are output as images.

[1236] Step 4:

[1237] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. For example, text can be sent using an API, and the results can be saved in text format.

[1238] python

[1239] import openai

[1240] openai.api_key = 'your-api-key'

[1241] response = openai.Completion.create(

[1242] engine="text-davinci-003",

[1243] prompt="Analyze and summarize the following text:\n" + all_text,

[1244] max_tokens=500

[1245] )

[1246] summary = response.choices[0].text

[1247] Preservation process

[1248] Step 5:

[1249] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts graph elements and features from each image and saves information such as model parameters and activation functions in text format.

[1250] python

[1251] from yolov5 import YOLOv5

[1252] yolo = YOLOv5("path / to / yolov5 / weights")

[1253] for img_path in image_save_paths:

[1254] results = yolo.predict(img_path)

[1255] parsed_results = results.pandas().xyxy Save process

[1256] Step 6:

[1257] The server extracts mathematical formulas and analyzes them using a mathematical formula analysis tool (e.g., Mathpix). OCR technology is used to convert the mathematical formula images into text format, and the resulting text data is saved in a format that can be understood.

[1258] python

[1259] import pytesseract

[1260] Import Image from PIL

[1261] math_images = [] Assumes that the images of mathematical formulas have been identified and saved in advance.

[1262] math_texts = [pytesseract.image_to_string(Image.open(img_path), config="math_configuration") for img_path in math_images]

[1263] Interpretation and storage

[1264] Step 7:

[1265] The server uses an emotion engine to recognize the user's emotions in real time. Emotional data is acquired based, for example, on user input, interactions, and video analysis from the webcam. Once emotional data is acquired, the analysis method and the way results are provided are adjusted accordingly.

[1266] python

[1267] import emotion_recognition

[1268] emotions = emotion_recognition.recognize_webcam()

[1269] Preservation process

[1270] Step 8:

[1271] The server integrates the analysis results of text, images, and mathematical formulas to generate the program code necessary for reproduction experiments. It combines the results of natural language processing models, image analysis, and mathematical formula analysis to generate Python code that enables reproduction experiments. In addition, it adds guides and explanations tailored to the user's emotions.

[1272] Step 9:

[1273] The server provides the user with the generated analysis results and program code. The method of delivery is adjusted according to the user's preferences; for example, a detailed step-by-step tutorial may be provided. The user then uses this to perform reproduction experiments.

[1274] (Example 2)

[1275] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[1276] In modern research, it is crucial to quickly and accurately understand the content of research papers and to conduct replication experiments. However, manually analyzing a vast number of papers and extracting the information necessary for summarization and replication experiments is an extremely time-consuming task. Furthermore, providing information that disregards the user's feelings is ineffective in understanding and utilizing the analysis results. This makes it difficult for researchers to conduct replication experiments efficiently, leading to a slowdown in research progress.

[1277] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[1278] In this invention, the server includes means for receiving electronic documents from a user, means for extracting document information, image information, and mathematical formula information from the electronic documents, means for using a natural language processing model to analyze the extracted document information, means for using an image analysis model to analyze the extracted image information, means for using a mathematical formula analysis tool to analyze the extracted mathematical formula information, means for integrating the analysis results and generating program code necessary for experimental reproduction, means for recognizing the user's emotions and adjusting the method and order of presentation of the analysis results, and means for providing the user with the analysis results and the generated program code. This makes it possible to quickly and accurately analyze the content of a paper and efficiently provide the information necessary for reproduction experiments. By providing information in accordance with the user's emotions, it is expected that the understanding of the analysis results will be enhanced and the reproduction experiments will proceed smoothly.

[1279] A "user" is an individual or organization that uses the system to upload electronic documents and receives analysis results and program code.

[1280] An "electronic document" refers to documents such as academic papers and research materials that are saved in PDF format or other digital formats.

[1281] "Document information" refers to text data extracted from electronic documents.

[1282] "Image information" refers to image data extracted from electronic documents.

[1283] "Mathematical formula information" refers to mathematical formula data extracted from electronic documents.

[1284] A "natural language processing model" is a machine learning model or algorithm used to analyze extracted document information and generate summaries and key points.

[1285] An "image analysis model" is a machine learning model or algorithm used to analyze extracted image information and extract the constituent elements and features of a figure or chart.

[1286] A "mathematical formula analysis tool" is software or algorithms that analyze extracted mathematical formula information and generate the meaning and application examples of the formulas in text format.

[1287] "Analysis results" refer to the results obtained using natural language processing models, image analysis models, and mathematical analysis tools.

[1288] "Program code" refers to the source code generated based on the analysis results, used to execute the reproduction experiment.

[1289] "Emotion recognition" refers to the process and technology of analyzing a user's emotions and adjusting the way results are presented and the order in which they are presented.

[1290] This invention is a system that analyzes electronic documents and comprehensively understands document information, image information, and mathematical formula information. This system allows users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system primarily operates on the server side, analyzing electronic documents uploaded by users and providing the results.

[1291] The server performs the following process: First, the user uploads an electronic document via the web interface. Then, the server receives this electronic document and temporarily stores it in storage. A possible specific storage location might be / tmp / uploaded_document.pdf.

[1292] Next, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information. Each page is analyzed separately, and the text is saved as a single continuous document, while images and mathematical formulas are saved in separate directories (e.g., / tmp / images / ).

[1293] The extracted document information is analyzed using a natural language processing model (e.g., GPT-4). This analysis generates a summary and key points of the paper, and the results are saved in text format (e.g., / tmp / summary.txt). Similarly, the extracted image information is analyzed using an image analysis model (e.g., YOLO, CNN), and the information obtained from the images is saved in text format (e.g., / tmp / image_analysis.txt). In addition, mathematical formula information is analyzed using a mathematical formula analysis tool (e.g., Mathpix), and the meaning and application examples of the formulas are described in natural language and saved as text data (e.g., / tmp / formula_analysis.txt).

[1294] Furthermore, the server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed in real time through analysis of the user's past interactions and video from the webcam (if permitted). It determines whether the user is showing anxiety or interest and stores the results in an internal data structure.

[1295] These analysis results are integrated to generate the program code necessary for reproduction experiments. The generated code includes the model definition, data preprocessing, and training procedures. For example, the generated code is saved in / tmp / reproduction_code.py. During this process, the server adjusts the order in which results are presented and the level of detail in the explanations based on the user's emotions. For example, if the user is feeling anxious, it provides detailed explanations and step-by-step guidance.

[1296] Finally, the server provides the user with the generated analysis results and program code. The user can download these from the web interface and use them as reference information for conducting reproduction experiments.

[1297] As a concrete example of operation, the following prompt statements are possible:

[1298] "I have uploaded a PDF of a paper on a new gradient descent method. Please analyze the contents of this PDF. Generate a summary of the text, images, and mathematical formulas, and then create Python code for a replication experiment based on these. Also, please adjust the presentation method of the analysis results based on my interests and concerns."

[1299] As described above, this system helps advance research activities by reducing the burden on users and efficiently providing analysis results.

[1300] The flow of the specific processing in Example 2 will be explained using Figure 13.

[1301] Step 1:

[1302] Users upload electronic documents (PDF files) to the server via a web interface.

[1303] Specifically, the user opens a browser and clicks the file upload button. They select a PDF file from the file selection dialog and click the "Upload" button.

[1304] Input: PDF file uploaded by the user

[1305] Output: PDF file sent to the server

[1306] Step 2:

[1307] The server receives the uploaded PDF file and temporarily stores it in storage.

[1308] Specifically, the file is saved to / tmp / uploaded_document.pdf.

[1309] Input: PDF file sent by the user to the server

[1310] Output: PDF file saved in the specified directory on the server.

[1311] Step 3:

[1312] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information.

[1313] Specifically, each page is analyzed, the text is saved as a continuous document, and images and mathematical formulas are saved in a separate directory (e.g., / tmp / images / ).

[1314] Input: Saved PDF file

[1315] Output: A dataset of extracted text, images, and mathematical formulas.

[1316] Step 4:

[1317] The server uses a natural language processing model (e.g., GPT-4) to analyze the extracted document information.

[1318] Specifically, the extracted text is input into a text analysis model, which then generates a summary and key points. The generated analysis results are saved in text format to / tmp / summary.txt.

[1319] Input: Extracted text data

[1320] Output: A text file containing summaries and key points.

[1321] Step 5:

[1322] The server analyzes image information using an image analysis model (e.g., YOLO, CNN).

[1323] Specifically, the image analysis model is run using each image file as input, and the information obtained from the analysis is saved to / tmp / image_analysis.txt.

[1324] Input: Extracted image data

[1325] Output: Text file containing image analysis results

[1326] Step 6:

[1327] The server uses a mathematical analysis tool (e.g., Mathpix) to analyze the mathematical information.

[1328] Specifically, the tool takes an image of a mathematical formula as input and converts it into text format. It then describes the meaning and application examples of the formula in natural language and saves them to / tmp / formula_analysis.txt.

[1329] Input: Extracted formula data

[1330] Output: A text file containing the results of the mathematical formula analysis.

[1331] Step 7:

[1332] The server uses an emotion engine to recognize the user's emotions.

[1333] Specifically, the system analyzes user input, interaction data, and, if permitted, webcam footage to determine whether the user is showing anxiety or interest. The results are stored in an internal data structure.

[1334] Input: User interaction data and video data

[1335] Output: Data on the user's emotional state

[1336] Step 8:

[1337] The server integrates the analysis results so far and generates the program code necessary for the reproduction experiment.

[1338] Specifically, the system combines the results of text analysis, image analysis, and mathematical analysis into a single file. It then incorporates a summary of the natural language processing model to generate Python code that allows for reproduction experiments. The generated code is saved in ` / tmp / reproduction_code.py`. Furthermore, the system adjusts the presentation order and level of detail of the results based on the user's emotions.

[1339] Input: Text analysis results, image analysis results, mathematical formula analysis results, user emotional state data

[1340] Output: Python code for reproduction experiment

[1341] Step 9:

[1342] The server provides the user with the generated analysis results and program code.

[1343] Specifically, the analysis results and code are presented as download links on a web interface. Users download them and perform reproduction experiments.

[1344] Input: Generated analysis results and program code

[1345] Output: Analysis results and program code provided to the user.

[1346] (Application Example 2)

[1347] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[1348] Currently, it is difficult for workers in logistics centers to quickly understand information about new technologies and algorithms and perform tasks efficiently based on that understanding. Furthermore, because there is no system in place to understand and appropriately respond to workers' emotions, anxiety and stress increase when the learning curve is steep, leading to decreased work efficiency. Therefore, there is a need to develop a system that can quickly and accurately analyze the content of research papers and adjust the method of information delivery according to the workers' emotions.

[1349] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for receiving a PDF of a paper from a user, means for extracting text, images, and mathematical formulas from the PDF, means for using a natural language processing model to analyze the extracted text, means for using an image analysis model to analyze the extracted images, means for using a mathematical formula analysis tool to analyze the extracted mathematical formulas, means for recognizing the user's emotions, means for integrating the analysis results and generating program code necessary for reproduction experiments, means for adjusting the method and presentation order of the analysis results based on the user's emotions, and means for providing the user with the analysis results and the generated program code. This enables an assist system that allows workers in a logistics center to quickly understand new technical information and perform work efficiently.

[1350] A "user" is a person or entity that uses the system to upload a PDF of a research paper and receive the analysis results.

[1351] A "paper" is a document that describes the results of academic research or studies in PDF format.

[1352] "PDF" stands for Portable Document Format, a file format that can contain text, images, and mathematical formulas in a document.

[1353] "Text" refers to the written information contained in a PDF, including sentences, explanations, and comments.

[1354] An "image" is an element that represents visual information such as graphs, diagrams, and photographs included within a PDF document.

[1355] "Mathematical formulas" refer to mathematical expressions written within a PDF document, including calculation formulas and equations.

[1356] A "natural language processing model" is an artificial intelligence technology used to analyze, summarize, and interpret text, such as a model that understands the meaning of extracted text.

[1357] An "image analysis model" is an artificial intelligence technology used to analyze images; for example, it is a model used to extract and understand information contained in an image.

[1358] A "mathematical formula analysis tool" is a tool used to analyze mathematical formulas and understand their meaning and application examples.

[1359] "Emotion recognition" is a technology that analyzes and identifies a user's emotions, and is about understanding the user's state of mind.

[1360] "Analysis results" refer to information and content generated based on the analysis of text, images, and mathematical formulas extracted from the PDF.

[1361] "Program code" refers to computer language code that describes the procedures and operations necessary to carry out a reproduction experiment.

[1362] "Means for adjusting the method of provision and presentation order" refers to a function that changes the way analysis results are displayed and the order in which they are presented based on the user's emotions.

[1363] To implement this invention, it is necessary to build a system that includes the following steps. Specific hardware includes a head-mounted display (HMD) for user use, a server, and a webcam. Software used may include PDF parsing libraries (e.g., PyMuPDF or PDFMiner), OCR software (e.g., Pytesseract), natural language processing models (e.g., Hugging Face Transformers), emotion recognition engines (e.g., EmotionRecognizer), and image analysis models (e.g., YOLO or CNN).

[1364] First, the user uploads the PDF of their paper to the server via the HMD (Head-Mounted Display). The server receives the PDF file and temporarily stores it in storage. This file is later broken down into text, images, and mathematical formulas using a PDF analysis library. The text is extracted using OCR software, and the images are saved in a separate directory. The mathematical formulas are saved as images and analyzed using a mathematical formula analysis tool.

[1365] Next, the server uses a natural language processing model to analyze the extracted text and generate a summary and key points of the paper. An image analysis model is used to extract elements and features of figures and tables from images, and the resulting information is saved in text format. A mathematical formula analysis tool is used to obtain text information from mathematical formula images, explaining the meaning and application examples of the formulas.

[1366] The server also uses a webcam to recognize the user's emotions in real time. The emotion recognition engine determines emotions from the user's input, interactions, and facial expressions, and evaluates whether the user is showing anxiety or interest.

[1367] The server integrates these analysis results and automatically generates the Python code necessary for replication experiments. It adjusts the presentation order and detailed explanations of the analysis results based on the user's emotional state. For example, if the user is feeling anxious, it provides a detailed step-by-step guide to help them confidently replicate the experiment.

[1368] Finally, the generated analysis results and program code are provided to the user via the HMD. The user can then use this as a reference to perform reproduction experiments.

[1369] As a concrete example, when a worker at a logistics center uploads a paper on a new cargo placement algorithm, the analysis results are displayed on a head-mounted display. If the worker has any questions or concerns, an emotion recognition engine detects this and automatically provides a detailed step-by-step guide.

[1370] Examples of prompts include: "Analyze the key points of the latest research paper on logistics centers. A detailed explanation of specific cargo placement algorithms is required. Also, suggest ways to alleviate worker anxiety."

[1371] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[1372] Step 1:

[1373] The user uploads a PDF of their paper to the server via a head-mounted display. The input is the PDF file selected by the user, and the output is this PDF file being saved on the server. The server temporarily stores the file in storage and passes it on to the next process.

[1374] Step 2:

[1375] The server parses a PDF file and extracts text, images, and mathematical formulas. The input is a PDF file stored on the server, and the output is the extracted text, images, and mathematical formulas. Specifically, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to process each page, concatenating the text into a single continuous text file, saving the images in a separate directory, and saving the mathematical formulas as images.

[1376] Step 3:

[1377] The server uses a natural language processing model to analyze the extracted text. The input is the extracted text, and the output is a summarized text or key points. Specifically, the server uses a natural language processing model (e.g., Hugging Face Transformers) to summarize the main sections of a paper and extract the core of a new method.

[1378] Step 4:

[1379] The server uses an image analysis model to analyze the extracted images. The input is the extracted images, and the output is text data of the constituent elements and features derived from the images. Specifically, the server uses an image analysis model (e.g., YOLO or CNN) to analyze the components of graphs and charts from the images and extract parameters and activation functions for each layer.

[1380] Step 5:

[1381] The server uses a mathematical expression analysis tool to analyze the extracted mathematical formulas. The input is an image of the formula, and the output is an interpretation of the formula in text format. Specifically, the server uses a mathematical expression analysis tool (e.g., Mathpix) to convert the image of the formula into text format and explains the meaning and application examples of the formula.

[1382] Step 6:

[1383] The server uses a webcam to recognize the user's emotions. The input is the user's video captured by the webcam, and the output is the estimated emotion of the user. Specifically, the server uses an emotion recognition engine (e.g., EmotionRecognizer) to determine the emotion in real time from the user's facial expressions and interactions.

[1384] Step 7:

[1385] The server integrates the analysis results to date and generates the program code necessary for the replication experiment. Inputs include analysis results of text, images, and mathematical formulas, as well as user sentiment information. Outputs are Python code required for the replication experiment. Specifically, the server uses a generative AI model to integrate the results and automatically generates program code that enables the replication experiment.

[1386] Step 8:

[1387] The server adjusts the method and order in which analysis results are presented based on the user's emotions. Input is the user's emotion information and generated program code, while output is the analysis results and program code formatted according to the emotion. Specifically, the server adds detailed explanations and step-by-step guides based on the emotion recognition results.

[1388] Step 9:

[1389] The server provides the user with the generated analysis results and program code. The input is the adjusted analysis results and program code, and the output is the result displayed on the user's head-mounted display. Specifically, the server allows the user to download or view the analysis results and code via a web interface.

[1390] The specific processing unit 290 transmits the result of the specific processing to the headset terminal 314. In the headset terminal 314, the control unit 46A causes the speaker 240 and display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[1391] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1392] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and specific processing may also be performed by the headset terminal 314.

[1393] [Fourth Embodiment]

[1394] Figure 7 shows an example of the configuration of the data processing system 410 according to the fourth embodiment.

[1395] As shown in Figure 7, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1396] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1397] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication interface 44, and a controlled object 443. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, camera 42, and controlled object 443 are also connected to the bus 52.

[1398] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[1399] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[1400] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[1401] The controlled object 443 includes a display device, LEDs in the eyes, and motors that drive the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. Furthermore, the robot 414's facial expressions can also be expressed by controlling the illumination state of the LEDs in its eyes.

[1402] Figure 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Figure 8, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[1403] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1404] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1405] In robot 414, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[1406] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[1407] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. This system operates on a server side and analyzes PDFs uploaded by users, providing the results.

[1408] Program processing flow

[1409] 1. Receiving the PDF

[1410] Users upload their research papers as PDFs to the server via a web interface. This serves as the entry point to the system.

[1411] 2. Data Extraction

[1412] The server saves the received PDF file to a specific directory. Next, a PDF parsing library is used to extract text, images, and mathematical formulas from the PDF.

[1413] The text is read page by page, combined, and saved as a single continuous text file. Images are saved in a specific folder, and mathematical formulas are converted into text format using OCR technology.

[1414] 3. Text Analysis

[1415] The server uses a natural language processing model to analyze the extracted text. This generates a summary and key points of the paper.

[1416] 4. Image Analysis

[1417] The server uses an image analysis model to analyze images within the paper. It retrieves data points from graphs and figures contained within the images, as well as their interpretations, in text format.

[1418] 5. Formula Analysis

[1419] The server uses a mathematical analysis tool to analyze the extracted mathematical formulas. The analysis results are saved as text data, including interpretations of the formulas' meanings and applications.

[1420] 6. Support for Integration and Reproducibility Experiments

[1421] The server integrates the analysis results and generates code for replication experiments. This allows users to obtain specific procedures and program code to reproduce the experiment based on the content of the paper.

[1422] 7. Providing the results

[1423] The server returns the generated analysis results and program code to the user. The user can then use this to gain a deeper understanding of the paper's content and smoothly carry out replication experiments.

[1424] Specific example

[1425] As a concrete example, we can consider the following scenario.

[1426] 1. Receiving the PDF

[1427] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[1428] 2. Data Extraction

[1429] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[1430] 3. Text Analysis

[1431] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[1432] 4. Image Analysis

[1433] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[1434] 5. Formula Analysis

[1435] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[1436] 6. Support for Integration and Reproducibility Experiments

[1437] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[1438] 7. Providing the results

[1439] The server provides the user with a generated summary, image analysis results, mathematical formula interpretations, and code for replication experiments. The user can then use this information to reproduce the paper's experiments in their own environment.

[1440] As described above, the system of the present invention enables users to deeply understand the content of research papers and efficiently conduct reproduction experiments.

[1441] The following describes the processing flow.

[1442] Step 1:

[1443] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[1444] Step 2:

[1445] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[1446] Step 3:

[1447] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[1448] Step 4:

[1449] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. The analysis results are saved in text format and used for subsequent processing.

[1450] Step 5:

[1451] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts the constituent elements and features of each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[1452] Step 6:

[1453] The server extracts mathematical formulas and analyzes them using a mathematical analysis tool (e.g., Mathpix). The formula images are converted into text format, and the meaning and application examples of the formulas are described in natural language. This allows the content of the formulas to be saved as text data.

[1454] Step 7:

[1455] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[1456] Step 8:

[1457] The server provides the user with the generated analysis results and program code. The user downloads this from the web interface and uses it as a reference to conduct reproduction experiments.

[1458] (Example 1)

[1459] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[1460] Conventional technologies required advanced expertise and considerable time to comprehensively analyze the content of papers and documents and facilitate reproduction experiments. Furthermore, the process of integrating results using multiple analysis methods and tools was complex and burdensome for users. Therefore, the present invention aims to provide a system that automates document analysis and preparation for reproduction experiments, enabling efficient and accurate execution.

[1461] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[1462] In this invention, the server includes means for receiving a document file from a user, means for extracting character data, figures, and formulas from the file, means for using a natural language processing model to analyze the extracted character data, means for using an image analysis model to analyze the extracted figures, means for using an analysis tool to analyze the extracted formulas, means for integrating the analysis results and generating a program necessary for reproduction verification, and means for providing the user with the analysis results and the generated program. This makes it possible to comprehensively analyze the contents of a document and to perform reproduction experiments efficiently and accurately.

[1463] A "user" refers to an entity that uploads document files to the server and receives the analysis results or programs.

[1464] "Document" refers to electronic files containing text information such as papers and reports, including formats such as PDF.

[1465] A "file" refers to digital data in which the contents of a document are stored.

[1466] "Character data" refers to the text information contained within a document.

[1467] "Graphics" refers to visual information such as images, graphs, and charts included within a document.

[1468] "Formula" refers to mathematical formulas or mathematical expressions included in a document.

[1469] "Extraction" refers to the process of taking specific data (text data, graphics, formulas) out of a file.

[1470] A "natural language processing model" refers to a machine learning model or algorithm used to analyze extracted text data and extract important information.

[1471] An "image analysis model" refers to a machine learning model or algorithm that analyzes extracted graphic data and converts visual information into text format.

[1472] An "analysis tool" refers to software or algorithms used to analyze extracted formulas and understand their meaning.

[1473] "Integration" refers to the process of combining analysis results into a single format and ensuring consistency.

[1474] "Reproduction verification" refers to the process of reproducing experiments or simulations based on the content of a document.

[1475] A "program" refers to a set of code and procedures necessary for reproducing and verifying a problem.

[1476] "Results" refers to the information obtained from the analysis of the document.

[1477] This invention is a document analysis and reproduction experiment support system, primarily configured to operate on the server side. The specific implementation methods for each step are described below.

[1478] Receiving document files

[1479] Users upload document files to the server using a web interface. These document files include PDFs and other formats containing text information such as papers and reports. The server saves the received document files to a specific directory (e.g., / tmp / pdf_uploads / ).

[1480] Data extraction

[1481] The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text data, shapes, and formulas from the file. The specific steps are as follows:

[1482] Text data: Extracts text from each page of a document and saves it as a series of text data.

[1483] Shapes: Extracts images, graphs, and charts from a document and saves them, for example, to a folder such as / tmp / pdf_uploads / images / .

[1484] Formula: Convert mathematical formulas to text format using an OCR library (e.g., Mathpix) and save them.

[1485] Data analysis

[1486] Analysis of character data

[1487] The server uses natural language processing models (e.g., BERT, GPT) to analyze text data and extract summaries and key points from papers. This allows the server to generate summaries that enable users to quickly understand the content of documents.

[1488] Analysis of geometric shapes

[1489] The server analyzes the shapes using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format. The analysis results include, for example, interpretations of the model's architecture diagram and experimental results graphs.

[1490] Analysis of the formula

[1491] The server uses a mathematical analysis tool (e.g., SymPy) to analyze mathematical formulas and generate interpretations of their meaning and applications. The analysis results are saved as text data explaining the meaning of the formulas.

[1492] Support for integration and replication experiments

[1493] The server integrates these analysis results and generates a program for replication experiments. This program is provided as Python code that includes model definition, data preprocessing, and training procedures. Users can use this program to replicate the experiment in their own environment.

[1494] Providing results

[1495] The server returns the generated analysis results and program code to the user. The analysis results include a summary, image analysis results, interpretation of mathematical formulas, and code for reproduction experiments, enabling the user to gain a deeper understanding of the document's content and efficiently conduct reproduction experiments.

[1496] Specific example

[1497] As a concrete example, consider a case where a user uploads a paper on a new learning method for AI models. The server can perform the analysis through the following steps:

[1498] The server extracts 40 pages of text, 5 images, and 3 mathematical formulas, and analyzes each of them.

[1499] The server uses a natural language processing model to generate a summary that reads, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[1500] The server uses an image analysis model to analyze the image, concluding that "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer."

[1501] The server uses a mathematical analysis tool to interpret Equation 1 as "showing the loss function of the proposed method, which minimizes the squared error."

[1502] The server generates Python code for reproduction experiments and provides it to the user.

[1503] Examples of prompt statements

[1504] "Analyze a paper on a new AI model learning method, extract key points, interpret image content and mathematical formulas, and generate code for replication experiments."

[1505] "Please analyze the following PDF paper and provide the summary, data points for the graphs, meaning of the formulas, and experimental reproduction procedures. PDF link: [link]"

[1506] As described above, the document analysis system of the present invention can provide an effective means for users to deeply understand the content of documents and efficiently conduct reproduction experiments.

[1507] The flow of the specific processing in Example 1 will be explained using Figure 11.

[1508] Step 1:

[1509] PDF reception

[1510] Input: The PDF file to be uploaded by the user.

[1511] Specific operation: On the web interface, the user clicks the "Select File" button, selects the desired PDF file from their local disk, and clicks the upload button.

[1512] Data processing: The server saves the received PDF files to a specific directory (e.g., / tmp / pdf_uploads / ).

[1513] Output: PDF file saved in / tmp / pdf_uploads / .

[1514] Step 2:

[1515] Data extraction

[1516] Input: Saved PDF file.

[1517] Specific operation: The server uses a PDF parsing library (e.g., PyMuPDF, pdftotext) to extract text, images, and mathematical formulas from PDF files.

[1518] Text extraction: Reads text from each page and saves it as a series of text files (e.g., / tmp / pdf_uploads / extracted_text.txt).

[1519] Image extraction: Identifies images within the PDF and saves them to a specific folder (e.g., / tmp / pdf_uploads / images / ).

[1520] Formula extraction: Use an OCR library (e.g., Mathpix) to convert formulas in the PDF into text format and save it (e.g., / tmp / pdf_uploads / extracted_equations.txt).

[1521] Data processing: Extraction of text, images, and mathematical formulas using a PDF analysis library.

[1522] Output: Text data ( / tmp / pdf_uploads / extracted_text.txt), image data ( / tmp / pdf_uploads / images / ), mathematical formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[1523] Step 3:

[1524] Text analysis

[1525] Input: Extracted text data ( / tmp / pdf_uploads / extracted_text.txt).

[1526] Specific operation: The server uses a natural language processing model (e.g., BERT, GPT) to analyze the text data and extract summaries and key points.

[1527] Section Identification: Analyzes text data to identify sections of a paper (e.g., Introduction, Methods, Results, Discussion, etc.).

[1528] Summary generation: Generate summaries for each section and extract the core elements of the new learning method.

[1529] Data processing: Text analysis and summarization using natural language processing models.

[1530] Output: Summarized text data.

[1531] Step 4:

[1532] Image analysis

[1533] Input: Extracted image data ( / tmp / pdf_uploads / images / ).

[1534] Specific operation: The server analyzes images using image analysis models (e.g., YOLO, OpenCV) and extracts important data points and interpretations in text format.

[1535] Graph and chart analysis: Extracting data points and interpretations from images (e.g., "Figure 1 shows a 3-layer convolutional neural network, with ReLU used in each layer").

[1536] Data processing: Analysis of image data using image analysis models.

[1537] Output: Text data of the image analysis results.

[1538] Step 5:

[1539] Formula analysis

[1540] Input: Extracted formula data ( / tmp / pdf_uploads / extracted_equations.txt).

[1541] Specific operation: The server uses a mathematical expression analysis tool (e.g., SymPy) to analyze the formula and generate an interpretation of its meaning and application.

[1542] Mathematical analysis: Analyze the mathematical formula and interpret it as "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[1543] Data calculation: Analysis of mathematical formulas using mathematical analysis tools.

[1544] Output: Text data of the mathematical analysis results.

[1545] Step 6:

[1546] Support for integration and replication experiments

[1547] Input: Text analysis results, image analysis results, mathematical formula analysis results.

[1548] Specific operation: The server integrates these analysis results and generates a program for reproduction experiments.

[1549] Integration of analysis results: Organize the analysis results and compile them into a single document (e.g., Markdown format).

[1550] Program generation: Generates Python code for replication experiments, describing the model definition, data preprocessing, and training procedures.

[1551] Data processing: Integration of analysis results and generation of program code.

[1552] Output: Integrated analysis results document and Python code.

[1553] Step 7:

[1554] Providing results

[1555] Input: Integrated analysis results document and Python code.

[1556] Specific operation: The server compresses the generated analysis results and program code into a zip file and provides a download link for the user.

[1557] Output: A download link that the user can access.

[1558] The above describes the specific operation and input / output of each processing step in this system.

[1559] (Application Example 1)

[1560] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[1561] When new production technologies or methods are proposed within a factory, actually reproducing and verifying them requires time and effort. Furthermore, understanding technical documents such as research papers requires specialized knowledge, and appropriately interpreting their content and reflecting it in experiments is difficult. As a result, the introduction of new technologies is often delayed, hindering improvements in production efficiency and quality.

[1562] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[1563] In this invention, the server includes means for receiving a PDF of a technical document from a user; means for extracting information, images, and mathematical representations from the PDF; means for using a natural language processing model to analyze the extracted information; means for using an image processing model to analyze the extracted images; means for using a mathematical expression analysis tool to analyze the extracted mathematical representations; means for integrating the analysis results and generating program code and automated experimental procedures necessary for reproduction experiments; and means for providing the analysis results and generated program code to the user. This makes it possible to quickly analyze new technologies described in technical documents within a factory and to automatically perform reproduction experiments.

[1564] A "user" is an individual or organization that operates the system to analyze technical documents or conduct reproduction experiments.

[1565] "Technical documents" refer to documents containing content related to science, technology, or engineering, such as research papers and technical reports.

[1566] "PDF" is an abbreviation for Portable Document Format, a file format for electronically displaying the contents of a document.

[1567] "Information" refers collectively to the text, images, and mathematical formulas included in technical documents.

[1568] "Images" refer to visual data such as diagrams, graphs, and photographs included within technical documents.

[1569] "Mathematical expression" refers to mathematical formulas and statistical descriptions included in technical documents.

[1570] A "natural language processing model" is a computational model used to analyze text data and understand its meaning and structure.

[1571] An "image processing model" is a computational model used to analyze image data and extract its content and features.

[1572] A "mathematical formula analysis tool" is software used to analyze mathematical formulas and understand their meaning and structure.

[1573] "Analysis results" refers to the collective data obtained by natural language processing models, image processing models, and mathematical analysis tools.

[1574] "Program code" refers to a computer program that describes the procedures and processes necessary to carry out a reproduction experiment.

[1575] An "automated experimental procedure" is a specific set of steps generated based on analysis results, which are used to perform a reproducible experiment using automated equipment.

[1576] A "system" is a general term for a complex set of devices or software that automates everything from the analysis of technical documents to reproduction experiments.

[1577] This invention provides a system that analyzes PDF technical documents and comprehensively analyzes text, images, and mathematical representations to facilitate users in performing replication experiments. This system includes a paper decoding system, a server, a cloud server, and a factory robot.

[1578] System Overview

[1579] Program processing

[1580] The server first receives technical document PDFs from users. The received PDFs are saved in a specific directory. Next, a PDF parsing library (e.g., PyPDF2) is used to extract text, images, and mathematical representations from the PDFs. The extracted text is analyzed using a natural language processing model (e.g., spacy) to generate summaries and key points of the papers. Images are analyzed using an image processing model (e.g., pytesseract) to convert the included data and graphs into text format. Mathematical representations are analyzed using a mathematical expression analysis tool (e.g., regular expressions) to explain their meaning.

[1581] The analysis results are integrated to generate the program code and automated experimental procedures necessary for reproduction experiments. This generated code is provided to the user. Furthermore, the system provides procedures and tools for automatically performing reproduction experiments based on the analysis results.

[1582] Hardware and software used

[1583] Hardware:

[1584] Factory robots (e.g., KUKA robots, ABB robots)

[1585] Cloud servers (AWS, Google Cloud, etc.)

[1586] software:

[1587] PDF analysis: PyPDF2

[1588] Natural Language Processing: Spacy

[1589] Image analysis: pytesseract

[1590] Formula extraction: Regular expressions (re)

[1591] Specific example

[1592] As a concrete example, consider a scenario in which a technical document on a new, highly efficient production technology is introduced to a factory. First, the factory manager uploads the PDF of the technical document to a factory robot using a smartphone. Analysis is performed on a cloud server, and the following results are generated.

[1593] Summary: "This technical document proposes a new technology that achieves a 20% improvement in efficiency compared to conventional technology."

[1594] Image analysis: "Figure 1 shows a flowchart of the new technology, and the main steps are as follows..."

[1595] Mathematical analysis: "Equation 1 shows the optimization parameters for improving efficiency."

[1596] Based on these analysis results, detailed program code necessary for reproduction experiments is generated. This generated code is provided to the factory robot, which automatically executes the reproduction experiment. This process allows for rapid verification of the effectiveness of new technologies and supports their implementation within the factory.

[1597] Example of a prompt

[1598] Examples of specific prompt messages are as follows:

[1599] "Please write a program for industrial robots to analyze papers on new production technologies and automate the replication experiments."

[1600] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[1601] Step 1:

[1602] PDF reception

[1603] The user uploads a PDF file of a technical document. This PDF is received by the server and saved to a specific directory.

[1604] Input: PDF file of a technical document.

[1605] Output: PDF file saved in the server directory.

[1606] Specific operation: When a user selects a PDF from the web interface and presses the upload button, the server saves the file to the specified directory.

[1607] Step 2:

[1608] Data extraction

[1609] The server uses the PDF parsing library (PyPDF2) to extract text, images, and mathematical representations from PDF files.

[1610] Input: Saved PDF file.

[1611] Output: Extracted text files, a list of image files, and a list of mathematical representations.

[1612] Specific operation: The server opens the PDF file, extracts text from each page, saves the image files, and adds the mathematical representations to a list.

[1613] Step 3:

[1614] Text analysis

[1615] The server uses a natural language processing model (SPacy) to analyze the extracted text.

[1616] Input: Extracted text file.

[1617] Output: Text summary and key points.

[1618] Specific operation: The server inputs a text file into a natural language processing model and extracts a summary and important sections.

[1619] Step 4:

[1620] Image analysis

[1621] The server uses an image processing model (pytesseract) to analyze the extracted images.

[1622] Input: List of extracted image files.

[1623] Output: Text data extracted from the image.

[1624] Specific operation: The server sequentially inputs each image file into the image processing model and extracts the text data.

[1625] Step 5:

[1626] Formula analysis

[1627] The server uses a mathematical expression analysis tool (regular expressions) to analyze the extracted mathematical representation.

[1628] Input: A list of mathematical expressions.

[1629] Output: Interpretation and explanation of the meaning of the formula.

[1630] Specific operation: The server inputs the list of mathematical formulas into a regular expression tool and generates text explaining the meaning of each formula.

[1631] Step 6:

[1632] Support for integration and replication experiments

[1633] The server integrates the analysis results and generates the program code and automated experimental procedures necessary for reproduction experiments.

[1634] Input: Text summaries, text data extracted from images, and interpretation of mathematical formulas.

[1635] Output: Integrated analysis results and program code for replication experiments.

[1636] Specific operation: The server integrates all data and generates Python code for reproduction experiments based on the analysis results.

[1637] Step 7:

[1638] Providing results

[1639] The server provides the user with the generated analysis results and program code.

[1640] Input: Integrated analysis results and program code for replication experiments.

[1641] Output: Analysis results and program code provided to the user.

[1642] Specific operation: The server formats the analysis results and code and provides them to the user as a download link. It also sends the analysis results via email or other methods as needed.

[1643] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[1644] This invention is a paper analysis system that analyzes PDF versions of research papers and comprehensively understands the content of text, images, and mathematical formulas, thereby enabling users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system operates on a server side, analyzing PDFs uploaded by users and providing the results.

[1645] Program processing flow

[1646] 1. Receiving the PDF

[1647] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the process begins when the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[1648] 2. Data Extraction

[1649] The server receives the PDF file and temporarily saves it to storage. For example, it might save it to / tmp / uploaded_paper.pdf and pass the file path to the next process.

[1650] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Each page is analyzed; text is concatenated into a single continuous text file, images are saved in a separate directory, and mathematical formulas are saved as images.

[1651] 3. Text Analysis

[1652] The server utilizes natural language processing models (e.g., GPT-4) to analyze the extracted text. This generates summaries and key points of the papers. The analysis results are saved in text format and used for subsequent processing.

[1653] 4. Image Analysis

[1654] The server performs the analysis using an image analysis model (e.g., YOLO, CNN). It extracts the components and features of the figures and tables from each image and saves the resulting information in text format. For example, it extracts parameters and activation functions for each layer.

[1655] 5. Formula Analysis

[1656] The server uses a mathematical formula analysis tool (e.g., Mathpix) to perform the analysis. It converts the mathematical formula image into text format and describes the meaning and application examples of the formula in natural language. This allows the content of the formula to be saved as text data.

[1657] 6. Emotion recognition

[1658] The server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed through user input, interaction, and video analysis from the webcam. For example, it can determine in real time whether the user is showing anxiety or interest.

[1659] 7. Support for Integration and Reproducibility Experiments

[1660] The server integrates the analysis results so far and generates the program code necessary for the replication experiment. It combines the results of the natural language processing model, image analysis, and mathematical analysis to generate Python code that enables the replication experiment.

[1661] Furthermore, the method of providing analysis results will be adjusted based on the user's emotions. For example, if a user is feeling anxious, more detailed explanations and step-by-step guides will be provided.

[1662] 8. Providing results

[1663] The server provides the user with the generated analysis results and program code. The order and method of presenting the results are adjusted according to the user's mood. For example, if the user appears interested, important results can be presented first to maintain their attention.

[1664] Users download this from the web interface and use it as a reference for conducting reproduction experiments. The reproduction experiment process proceeds based on the provided code and explanations.

[1665] Specific example

[1666] As a concrete example, we can consider the following scenario.

[1667] 1. Receiving the PDF

[1668] The user uploads a paper titled "A New Learning Method for AI Models" to the server. This PDF document details a new method in AI research.

[1669] 2. Data Extraction

[1670] The server extracts 40 pages of text, 5 images, and 3 complex mathematical formulas from the PDF. Each image includes graphs and model architecture diagrams.

[1671] 3. Text Analysis

[1672] The server uses a natural language processing model to summarize the main sections of the paper and extract the core of the new learning method. A summary is generated such as, "This paper proposes a new gradient descent method that converges faster than conventional methods."

[1673] 4. Image Analysis

[1674] The server extracts the parameters and activation functions of each layer from the image in text format. Information such as "Figure 1 shows a three-layer convolutional neural network, with ReLU used in each layer" can be obtained.

[1675] 5. Formula Analysis

[1676] The server analyzes the formula and generates text explaining its meaning. It interprets it as: "Formula 1 represents the loss function of the proposed method, which minimizes the squared error."

[1677] 6. Emotion recognition

[1678] The server recognizes the user's emotions and determines whether the user is feeling anxious about the analysis results. For example, if a user has asked many questions about past analysis results, the server will determine that the user is feeling anxious.

[1679] 7. Support for Integration and Reproducibility Experiments

[1680] The server generates Python code for the replication experiment based on this information. This code includes the model definition, data preprocessing, and training procedures.

[1681] At the same time, the order in which the analysis results are presented and the amount of detailed explanation are adjusted based on the user's emotions.

[1682] 8. Providing results

[1683] The server provides the user with generated summaries, image analysis results, mathematical formula interpretations, and code for reproduction experiments.

[1684] Users receive information adjusted through emotion recognition and can utilize the analysis results in a format that suits their own emotions, thereby facilitating the smooth progress of replication experiments.

[1685] The following describes the processing flow.

[1686] Step 1:

[1687] The user uploads the PDF of their paper to the server via a web interface. This is the entry point to the system; the user selects the PDF file using their browser's file upload function and clicks the "Upload" button.

[1688] Step 2:

[1689] The server receives the PDF file and temporarily saves it to storage. For example, the save location is set to / tmp / uploaded_paper.pdf, and the file path is checked when passing it to the next process.

[1690] Step 3:

[1691] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to open the PDF and extract text, images, and mathematical formulas from the paper. Text is extracted from each page and combined into a single continuous text file. Images are saved to a specific directory, and mathematical formulas are output as images.

[1692] Step 4:

[1693] The server analyzes the extracted text using a natural language processing model (e.g., GPT-4). This generates a summary of the paper and key points. For example, text can be sent using an API, and the results can be saved in text format.

[1694] python

[1695] import openai

[1696] openai.api_key = 'your-api-key'

[1697] response = openai.Completion.create(

[1698] engine="text-davinci-003",

[1699] prompt="Analyze and summarize the following text:\n" + all_text,

[1700] max_tokens=500

[1701] )

[1702] summary = response.choices[0].text

[1703] Preservation process

[1704] Step 5:

[1705] The server analyzes the extracted images using an image analysis model (e.g., YOLO, CNN). It extracts graph elements and features from each image and saves information such as model parameters and activation functions in text format.

[1706] python

[1707] from yolov5 import YOLOv5

[1708] yolo = YOLOv5("path / to / yolov5 / weights")

[1709] for img_path in image_save_paths:

[1710] results = yolo.predict(img_path)

[1711] parsed_results = results.pandas().xyxy Save process

[1712] Step 6:

[1713] The server extracts mathematical formulas and analyzes them using a mathematical formula analysis tool (e.g., Mathpix). OCR technology is used to convert the mathematical formula images into text format, and the resulting text data is saved in a format that can be understood.

[1714] python

[1715] import pytesseract

[1716] Import Image from PIL

[1717] math_images = [] Assumes that the images of mathematical formulas have been identified and saved in advance.

[1718] math_texts = [pytesseract.image_to_string(Image.open(img_path), config="math_configuration") for img_path in math_images]

[1719] Interpretation and storage

[1720] Step 7:

[1721] The server uses an emotion engine to recognize the user's emotions in real time. Emotional data is acquired based, for example, on user input, interactions, and video analysis from the webcam. Once emotional data is acquired, the analysis method and the way results are provided are adjusted accordingly.

[1722] python

[1723] import emotion_recognition

[1724] emotions = emotion_recognition.recognize_webcam()

[1725] Preservation process

[1726] Step 8:

[1727] The server integrates the analysis results of text, images, and mathematical formulas to generate the program code necessary for reproduction experiments. It combines the results of natural language processing models, image analysis, and mathematical formula analysis to generate Python code that enables reproduction experiments. In addition, it adds guides and explanations tailored to the user's emotions.

[1728] Step 9:

[1729] The server provides the user with the generated analysis results and program code. The method of delivery is adjusted according to the user's preferences; for example, a detailed step-by-step tutorial may be provided. The user then uses this to perform reproduction experiments.

[1730] (Example 2)

[1731] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[1732] In modern research, it is crucial to quickly and accurately understand the content of research papers and to conduct replication experiments. However, manually analyzing a vast number of papers and extracting the information necessary for summarization and replication experiments is an extremely time-consuming task. Furthermore, providing information that disregards the user's feelings is ineffective in understanding and utilizing the analysis results. This makes it difficult for researchers to conduct replication experiments efficiently, leading to a slowdown in research progress.

[1733] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[1734] In this invention, the server includes means for receiving electronic documents from a user, means for extracting document information, image information, and mathematical formula information from the electronic documents, means for using a natural language processing model to analyze the extracted document information, means for using an image analysis model to analyze the extracted image information, means for using a mathematical formula analysis tool to analyze the extracted mathematical formula information, means for integrating the analysis results and generating program code necessary for experimental reproduction, means for recognizing the user's emotions and adjusting the method and order of presentation of the analysis results, and means for providing the user with the analysis results and the generated program code. This makes it possible to quickly and accurately analyze the content of a paper and efficiently provide the information necessary for reproduction experiments. By providing information in accordance with the user's emotions, it is expected that the understanding of the analysis results will be enhanced and the reproduction experiments will proceed smoothly.

[1735] A "user" is an individual or organization that uses the system to upload electronic documents and receives analysis results and program code.

[1736] An "electronic document" refers to documents such as academic papers and research materials that are saved in PDF format or other digital formats.

[1737] "Document information" refers to text data extracted from electronic documents.

[1738] "Image information" refers to image data extracted from electronic documents.

[1739] "Mathematical formula information" refers to mathematical formula data extracted from electronic documents.

[1740] A "natural language processing model" is a machine learning model or algorithm used to analyze extracted document information and generate summaries and key points.

[1741] An "image analysis model" is a machine learning model or algorithm used to analyze extracted image information and extract the constituent elements and features of a figure or chart.

[1742] A "mathematical formula analysis tool" is software or algorithms that analyze extracted mathematical formula information and generate the meaning and application examples of the formulas in text format.

[1743] "Analysis results" refer to the results obtained using natural language processing models, image analysis models, and mathematical analysis tools.

[1744] "Program code" refers to the source code generated based on the analysis results, used to execute the reproduction experiment.

[1745] "Emotion recognition" refers to the process and technology of analyzing a user's emotions and adjusting the way results are presented and the order in which they are presented.

[1746] This invention is a system that analyzes electronic documents and comprehensively understands document information, image information, and mathematical formula information. This system allows users to easily perform replication experiments. It also includes a function to recognize the user's emotions and adjust the method and order of presentation of analysis results accordingly. This system primarily operates on the server side, analyzing electronic documents uploaded by users and providing the results.

[1747] The server performs the following process: First, the user uploads an electronic document via the web interface. Then, the server receives this electronic document and temporarily stores it in storage. A possible specific storage location might be / tmp / uploaded_document.pdf.

[1748] Next, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information. Each page is analyzed separately, and the text is saved as a single continuous document, while images and mathematical formulas are saved in separate directories (e.g., / tmp / images / ).

[1749] The extracted document information is analyzed using a natural language processing model (e.g., GPT-4). This analysis generates a summary and key points of the paper, and the results are saved in text format (e.g., / tmp / summary.txt). Similarly, the extracted image information is analyzed using an image analysis model (e.g., YOLO, CNN), and the information obtained from the images is saved in text format (e.g., / tmp / image_analysis.txt). In addition, mathematical formula information is analyzed using a mathematical formula analysis tool (e.g., Mathpix), and the meaning and application examples of the formulas are described in natural language and saved as text data (e.g., / tmp / formula_analysis.txt).

[1750] Furthermore, the server uses an emotion engine to recognize the user's emotions. Emotion recognition is performed in real time through analysis of the user's past interactions and video from the webcam (if permitted). It determines whether the user is showing anxiety or interest and stores the results in an internal data structure.

[1751] These analysis results are integrated to generate the program code necessary for reproduction experiments. The generated code includes the model definition, data preprocessing, and training procedures. For example, the generated code is saved in / tmp / reproduction_code.py. During this process, the server adjusts the order in which results are presented and the level of detail in the explanations based on the user's emotions. For example, if the user is feeling anxious, it provides detailed explanations and step-by-step guidance.

[1752] Finally, the server provides the user with the generated analysis results and program code. The user can download these from the web interface and use them as reference information for conducting reproduction experiments.

[1753] As a concrete example of operation, the following prompt statements are possible:

[1754] "I have uploaded a PDF of a paper on a new gradient descent method. Please analyze the contents of this PDF. Generate a summary of the text, images, and mathematical formulas, and then create Python code for a replication experiment based on these. Also, please adjust the presentation method of the analysis results based on my interests and concerns."

[1755] As described above, this system helps advance research activities by reducing the burden on users and efficiently providing analysis results.

[1756] The flow of the specific processing in Example 2 will be explained using Figure 13.

[1757] Step 1:

[1758] Users upload electronic documents (PDF files) to the server via a web interface.

[1759] Specifically, the user opens a browser and clicks the file upload button. They select a PDF file from the file selection dialog and click the "Upload" button.

[1760] Input: PDF file uploaded by the user

[1761] Output: PDF file sent to the server

[1762] Step 2:

[1763] The server receives the uploaded PDF file and temporarily stores it in storage.

[1764] Specifically, the file is saved to / tmp / uploaded_document.pdf.

[1765] Input: PDF file sent by the user to the server

[1766] Output: PDF file saved in the specified directory on the server.

[1767] Step 3:

[1768] The server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to analyze the PDF and extract document information, image information, and mathematical formula information.

[1769] Specifically, each page is analyzed, the text is saved as a continuous document, and images and mathematical formulas are saved in a separate directory (e.g., / tmp / images / ).

[1770] Input: Saved PDF file

[1771] Output: A dataset of extracted text, images, and mathematical formulas.

[1772] Step 4:

[1773] The server uses a natural language processing model (e.g., GPT-4) to analyze the extracted document information.

[1774] Specifically, the extracted text is input into a text analysis model, which then generates a summary and key points. The generated analysis results are saved in text format to / tmp / summary.txt.

[1775] Input: Extracted text data

[1776] Output: A text file containing summaries and key points.

[1777] Step 5:

[1778] The server analyzes image information using an image analysis model (e.g., YOLO, CNN).

[1779] Specifically, the image analysis model is run using each image file as input, and the information obtained from the analysis is saved to / tmp / image_analysis.txt.

[1780] Input: Extracted image data

[1781] Output: Text file containing image analysis results

[1782] Step 6:

[1783] The server uses a mathematical analysis tool (e.g., Mathpix) to analyze the mathematical information.

[1784] Specifically, the tool takes an image of a mathematical formula as input and converts it into text format. It then describes the meaning and application examples of the formula in natural language and saves them to / tmp / formula_analysis.txt.

[1785] Input: Extracted formula data

[1786] Output: A text file containing the results of the mathematical formula analysis.

[1787] Step 7:

[1788] The server uses an emotion engine to recognize the user's emotions.

[1789] Specifically, the system analyzes user input, interaction data, and, if permitted, webcam footage to determine whether the user is showing anxiety or interest. The results are stored in an internal data structure.

[1790] Input: User interaction data and video data

[1791] Output: Data on the user's emotional state

[1792] Step 8:

[1793] The server integrates the analysis results so far and generates the program code necessary for the reproduction experiment.

[1794] Specifically, the system combines the results of text analysis, image analysis, and mathematical analysis into a single file. It then incorporates a summary of the natural language processing model to generate Python code that allows for reproduction experiments. The generated code is saved in ` / tmp / reproduction_code.py`. Furthermore, the system adjusts the presentation order and level of detail of the results based on the user's emotions.

[1795] Input: Text analysis results, image analysis results, mathematical formula analysis results, user emotional state data

[1796] Output: Python code for reproduction experiment

[1797] Step 9:

[1798] The server provides the user with the generated analysis results and program code.

[1799] Specifically, the analysis results and code are presented as download links on a web interface. Users download them and perform reproduction experiments.

[1800] Input: Generated analysis results and program code

[1801] Output: Analysis results and program code provided to the user.

[1802] (Application Example 2)

[1803] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[1804] Currently, it is difficult for workers in logistics centers to quickly understand information about new technologies and algorithms and perform tasks efficiently based on that understanding. Furthermore, because there is no system in place to understand and appropriately respond to workers' emotions, anxiety and stress increase when the learning curve is steep, leading to decreased work efficiency. Therefore, there is a need to develop a system that can quickly and accurately analyze the content of research papers and adjust the method of information delivery according to the workers' emotions.

[1805] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for receiving a PDF of a paper from a user, means for extracting text, images, and mathematical formulas from the PDF, means for using a natural language processing model to analyze the extracted text, means for using an image analysis model to analyze the extracted images, means for using a mathematical formula analysis tool to analyze the extracted mathematical formulas, means for recognizing the user's emotions, means for integrating the analysis results and generating program code necessary for reproduction experiments, means for adjusting the method and presentation order of the analysis results based on the user's emotions, and means for providing the user with the analysis results and the generated program code. This enables an assist system that allows workers in a logistics center to quickly understand new technical information and perform work efficiently.

[1806] A "user" is a person or entity that uses the system to upload a PDF of a research paper and receive the analysis results.

[1807] A "paper" is a document that describes the results of academic research or studies in PDF format.

[1808] "PDF" stands for Portable Document Format, a file format that can contain text, images, and mathematical formulas in a document.

[1809] "Text" refers to the written information contained in a PDF, including sentences, explanations, and comments.

[1810] An "image" is an element that represents visual information such as graphs, diagrams, and photographs included within a PDF document.

[1811] "Mathematical formulas" refer to mathematical expressions written within a PDF document, including calculation formulas and equations.

[1812] A "natural language processing model" is an artificial intelligence technology used to analyze, summarize, and interpret text, such as a model that understands the meaning of extracted text.

[1813] An "image analysis model" is an artificial intelligence technology used to analyze images; for example, it is a model used to extract and understand information contained in an image.

[1814] A "mathematical formula analysis tool" is a tool used to analyze mathematical formulas and understand their meaning and application examples.

[1815] "Emotion recognition" is a technology that analyzes and identifies a user's emotions, and is about understanding the user's state of mind.

[1816] "Analysis results" refer to information and content generated based on the analysis of text, images, and mathematical formulas extracted from the PDF.

[1817] "Program code" refers to computer language code that describes the procedures and operations necessary to carry out a reproduction experiment.

[1818] "Means for adjusting the method of provision and presentation order" refers to a function that changes the way analysis results are displayed and the order in which they are presented based on the user's emotions.

[1819] To implement this invention, it is necessary to build a system that includes the following steps. Specific hardware includes a head-mounted display (HMD) for user use, a server, and a webcam. Software used may include PDF parsing libraries (e.g., PyMuPDF or PDFMiner), OCR software (e.g., Pytesseract), natural language processing models (e.g., Hugging Face Transformers), emotion recognition engines (e.g., EmotionRecognizer), and image analysis models (e.g., YOLO or CNN).

[1820] First, the user uploads the PDF of their paper to the server via the HMD (Head-Mounted Display). The server receives the PDF file and temporarily stores it in storage. This file is later broken down into text, images, and mathematical formulas using a PDF analysis library. The text is extracted using OCR software, and the images are saved in a separate directory. The mathematical formulas are saved as images and analyzed using a mathematical formula analysis tool.

[1821] Next, the server uses a natural language processing model to analyze the extracted text and generate a summary and key points of the paper. An image analysis model is used to extract elements and features of figures and tables from images, and the resulting information is saved in text format. A mathematical formula analysis tool is used to obtain text information from mathematical formula images, explaining the meaning and application examples of the formulas.

[1822] The server also uses a webcam to recognize the user's emotions in real time. The emotion recognition engine determines emotions from the user's input, interactions, and facial expressions, and evaluates whether the user is showing anxiety or interest.

[1823] The server integrates these analysis results and automatically generates the Python code necessary for replication experiments. It adjusts the presentation order and detailed explanations of the analysis results based on the user's emotional state. For example, if the user is feeling anxious, it provides a detailed step-by-step guide to help them confidently replicate the experiment.

[1824] Finally, the generated analysis results and program code are provided to the user via the HMD. The user can then use this as a reference to perform reproduction experiments.

[1825] As a concrete example, when a worker at a logistics center uploads a paper on a new cargo placement algorithm, the analysis results are displayed on a head-mounted display. If the worker has any questions or concerns, an emotion recognition engine detects this and automatically provides a detailed step-by-step guide.

[1826] Examples of prompts include: "Analyze the key points of the latest research paper on logistics centers. A detailed explanation of specific cargo placement algorithms is required. Also, suggest ways to alleviate worker anxiety."

[1827] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[1828] Step 1:

[1829] The user uploads a PDF of their paper to the server via a head-mounted display. The input is the PDF file selected by the user, and the output is this PDF file being saved on the server. The server temporarily stores the file in storage and passes it on to the next process.

[1830] Step 2:

[1831] The server parses a PDF file and extracts text, images, and mathematical formulas. The input is a PDF file stored on the server, and the output is the extracted text, images, and mathematical formulas. Specifically, the server uses a PDF parsing library (e.g., PyMuPDF or PDFMiner) to process each page, concatenating the text into a single continuous text file, saving the images in a separate directory, and saving the mathematical formulas as images.

[1832] Step 3:

[1833] The server uses a natural language processing model to analyze the extracted text. The input is the extracted text, and the output is a summarized text or key points. Specifically, the server uses a natural language processing model (e.g., Hugging Face Transformers) to summarize the main sections of a paper and extract the core of a new method.

[1834] Step 4:

[1835] The server uses an image analysis model to analyze the extracted images. The input is the extracted images, and the output is text data of the constituent elements and features derived from the images. Specifically, the server uses an image analysis model (e.g., YOLO or CNN) to analyze the components of graphs and charts from the images and extract parameters and activation functions for each layer.

[1836] Step 5:

[1837] The server uses a mathematical expression analysis tool to analyze the extracted mathematical formulas. The input is an image of the formula, and the output is an interpretation of the formula in text format. Specifically, the server uses a mathematical expression analysis tool (e.g., Mathpix) to convert the image of the formula into text format and explains the meaning and application examples of the formula.

[1838] Step 6:

[1839] The server uses a webcam to recognize the user's emotions. The input is the user's video captured by the webcam, and the output is the estimated emotion of the user. Specifically, the server uses an emotion recognition engine (e.g., EmotionRecognizer) to determine the emotion in real time from the user's facial expressions and interactions.

[1840] Step 7:

[1841] The server integrates the analysis results to date and generates the program code necessary for the replication experiment. Inputs include analysis results of text, images, and mathematical formulas, as well as user sentiment information. Outputs are Python code required for the replication experiment. Specifically, the server uses a generative AI model to integrate the results and automatically generates program code that enables the replication experiment.

[1842] Step 8:

[1843] The server adjusts the method and order in which analysis results are presented based on the user's emotions. Input is the user's emotion information and generated program code, while output is the analysis results and program code formatted according to the emotion. Specifically, the server adds detailed explanations and step-by-step guides based on the emotion recognition results.

[1844] Step 9:

[1845] The server provides the user with the generated analysis results and program code. The input is the adjusted analysis results and program code, and the output is the result displayed on the user's head-mounted display. Specifically, the server allows the user to download or view the analysis results and code via a web interface.

[1846] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[1847] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1848] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the robot 414.

[1849] Furthermore, the emotion identification model 59, acting as an emotion engine, may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to a specific mapping, which is an emotion map (see Figure 9). Similarly, the emotion identification model 59 may also determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1850] Figure 9 shows an emotion map 400 in which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. The closer to the center of the concentric circles, the more primitive the emotions are located. Further out of the concentric circles, emotions representing states and actions arising from mental states are located. Emotion is a concept that includes feelings and mental states. On the left side of the concentric circles, emotions that are generally generated from reactions occurring in the brain are located. On the right side of the concentric circles, emotions that are generally induced by situational judgment are located. Above and below the concentric circles, emotions that are generally generated from reactions occurring in the brain and induced by situational judgment are located. In addition, the emotion of "pleasure" is located on the upper side of the concentric circles, and the emotion of "displeasure" is located on the lower side. Thus, in the emotion map 400, multiple emotions are mapped based on the structure in which emotions arise, and emotions that are likely to occur simultaneously are mapped close together.

[1851] These emotions are distributed at the 3 o'clock position on the Emotion Map 400, and usually fluctuate between feelings of security and anxiety. In the right half of the Emotion Map 400, situational awareness takes precedence over internal feelings, resulting in a calm impression.

[1852] The inside of the Emotion Map 400 represents inner thoughts, while the outside represents actions. Therefore, the further you go from the outside of the Emotion Map 400, the more visible (expressed in actions) your emotions become.

[1853] Here, human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, it results in discomfort, and when they approach the ideal, it results in pleasure. Similarly, in robots, cars, motorcycles, etc., emotions can be created based on various balances, such as posture and battery level. When these balances deviate from the ideal, it results in discomfort, and when they approach the ideal, it results in pleasure. The emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on a system for analyzing brain physiological signals of speech emotion recognition and emotion, Tokushima University, doctoral dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map contains emotions belonging to a region called "response," where sensation is dominant. The right half of the emotion map contains emotions belonging to a region called "situation," where situational awareness is dominant.

[1854] The emotion map defines two emotions that promote learning. One is the emotion around the middle of the negative "repentance" and "reflection" on the situation side. In other words, it is when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is the emotion around the positive "desire" on the reaction side. In other words, it is when the robot has positive feelings such as "I want more" or "I want to know more."

[1855] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​representing each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple training data sets, which are combinations of user input and emotion values ​​representing each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions located close together have similar values, as shown in the emotion map 900 in Figure 10. Figure 10 shows an example where multiple emotions such as "reassured," "calm," and "confident" have similar emotion values.

[1856] The above description primarily focuses on the functions of the data processing device 12 in relation to this disclosure. However, the system related to this disclosure is not necessarily implemented on a server. The system related to this disclosure may be implemented as a general information processing system. This disclosure may be implemented, for example, as a software program that runs on a personal computer or as an application that runs on a smartphone. The method related to this disclosure may be provided to users in SaaS (Software as a Service) format.

[1857] In the above embodiment, an example was given in which a specific process is performed by a single computer 22. However, the technology of this disclosure is not limited thereto, and a distributed processing of the specific process may be performed by multiple computers, including computer 22. For example, a data generation model 58 may be provided in an external device of the data processing device 12, and the external device may generate data according to the input data.

[1858] In the above embodiment, an example was given in which the specific processing program 56 is stored in the storage 32, but the technology of this disclosure is not limited thereto. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-temporary storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-temporary storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes specific processing according to the specific processing program 56.

[1859] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[1860] Furthermore, it is not necessary to store the entirety of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store the entirety of the specific processing program 56 in the storage 32; it is acceptable to store only a portion of the specific processing program 56.

[1861] The following types of processors can be used as hardware resources to perform specific processing. Examples of processors include a CPU, a general-purpose processor that functions as a hardware resource to perform specific processing by executing software, i.e., a program. Other examples of processors include dedicated electrical circuits, such as FPGAs (Field-Programmable Gate Arrays), PLDs (Programmable Logic Devices), or ASICs (Application Specific Integrated Circuits), which have circuit configurations specifically designed to perform specific processing. All of these processors have built-in or connected memory, and all of them perform specific processing by using memory.

[1862] The hardware resource that performs a specific process may consist of one of these various processors, or it may consist of a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Alternatively, the hardware resource that performs a specific process may consist of a single processor.

[1863] Examples of configurations using a single processor include, firstly, a configuration in which one or more CPUs and software are combined to form a single processor, and this processor functions as a hardware resource that performs a specific process. Secondly, there is a configuration using a processor that realizes the functions of the entire system, including multiple hardware resources that perform a specific process, on a single IC chip, as exemplified by SoCs (System-on-a-chip). In this way, a specific process is realized using one or more of the above types of processors as hardware resources.

[1864] Furthermore, the hardware structure of these various processors can more specifically utilize electrical circuits that combine circuit elements such as semiconductor devices. Also, the specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps can be deleted, new steps added, or the processing order rearranged, as long as it does not deviate from the main purpose.

[1865] The descriptions and illustrations presented above are detailed explanations of the technical aspects of this disclosure and are merely examples of the technical aspects. For example, the above descriptions of the structure, function, operation, and effect are examples of the structure, function, operation, and effect of the technical aspects of this disclosure. Therefore, it goes without saying that you may delete unnecessary parts, add new elements, or replace elements in the descriptions and illustrations presented above, as long as you do not deviate from the essence of the technical aspects of this disclosure. Furthermore, in order to avoid confusion and facilitate understanding of the technical aspects of this disclosure, explanations of common technical knowledge and the like that do not require special explanation to enable the implementation of the technical aspects of this disclosure have been omitted from the descriptions and illustrations presented above.

[1866] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

[1867] The following is further disclosed regarding the embodiments described above.

[1868] (Claim 1)

[1869] A means of receiving PDFs of papers from users,

[1870] A means for extracting text, images, and mathematical formulas from the aforementioned PDF,

[1871] A method using a natural language processing model to analyze the extracted text,

[1872] A means of using an image analysis model to analyze the extracted images,

[1873] One method is to use a mathematical analysis tool to analyze the extracted mathematical formulas,

[1874] A means for integrating the aforementioned analysis results and generating program code necessary for reproduction experiments,

[1875] Means for providing the user with the analysis results and the generated program code,

[1876] A system that includes this.

[1877] (Claim 2)

[1878] The system according to claim 1, which performs analysis on each page with respect to the extraction of the aforementioned text, images, and mathematical formulas.

[1879] (Claim 3)

[1880] The system according to claim 1, further comprising means for providing procedures and tools necessary for reproduction experiments based on the aforementioned analysis results.

[1881] "Example 1"

[1882] (Claim 1)

[1883] A means of receiving document files from users,

[1884] Means for extracting text data, figures, and formulas from the aforementioned file,

[1885] A method using a natural language processing model to analyze extracted character data,

[1886] A method using an image analysis model to analyze extracted shapes,

[1887] Methods include using an analysis tool to analyze the extracted formulas,

[1888] A means for integrating the aforementioned analysis results and generating a program necessary for reproduction verification,

[1889] Means for providing the user with the analysis results and the generated program,

[1890] A system that includes this.

[1891] (Claim 2)

[1892] The system according to claim 1, which performs analysis on each page with respect to the extraction of the aforementioned character data, figures, and formulas.

[1893] (Claim 3)

[1894] The system according to claim 1, further comprising means for providing procedures and tools necessary for reproduction verification based on the aforementioned analysis results.

[1895] "Application Example 1"

[1896] (Claim 1)

[1897] A means of receiving technical documents in PDF format from users,

[1898] A means for extracting information, images, and mathematical representations from the aforementioned PDF,

[1899] A method using a natural language processing model to analy...

Claims

1. A means of receiving PDFs of papers from users, A means for extracting text, images, and mathematical formulas from the aforementioned PDF, A method using a natural language processing model to analyze the extracted text, A means of using an image analysis model to analyze the extracted images, One method is to use a mathematical analysis tool to analyze the extracted mathematical formulas, A means for integrating the aforementioned analysis results and generating program code necessary for reproduction experiments, Means for providing the user with the analysis results and the generated program code, A system that includes this.

2. The system according to claim 1, which performs analysis on each page with respect to the extraction of the aforementioned text, images, and mathematical formulas.

3. The system according to claim 1, further comprising means for providing procedures and tools necessary for reproduction experiments based on the aforementioned analysis results.

Images