System

Abstract

A system is provided.SOLUTION: The system includes means for acquiring images and slides from past materials, means for extracting and vectorizing the acquired images and slides as feature amounts, means for receiving a search query input by a user as a text and vectorizing the search query, means for calculating a similarity between the vector of the search query and the vectors of the past materials and listing materials with high similarity, and means for displaying materials with high similarity.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, 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] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] When creating a project or proposal document, if a user wants to reuse an image or slide that they have seen in the past, they often cannot remember which document it was used in. Furthermore, with conventional search methods, users are unable to efficiently find the slides or images they need because they have a vague memory of the name of the service or project, or keywords that are likely to be used in the slides. To address this issue, a search method that allows users to search based on the atmosphere of the image or slide is needed. [Means for solving the problem]

[0005] The present invention is a system that includes a means for acquiring images and slides from past documents, a means for extracting features from the acquired images and slides and vectorizing them, a means for receiving a search query entered by a user as text and vectorizing it, a means for calculating the similarity between the vectorized search query and the vectors of past documents, a means for listing documents with high similarity, and a means for displaying documents with high similarity. This allows users to efficiently search for desired documents based on the atmosphere of the images and slides rather than specific keywords. The system also includes a feature that natural language processing technology is used to vectorize the text query and a deep learning model is used to extract features from the images and slides.

[0006] "Past materials" refers to electronic data such as planning materials and proposal materials that the user has created or saved in the past.

[0007] "Images and slides" refers to data containing visual information contained in past documents, which can be saved in formats such as JPEG, PNG, PDF, and PPT.

[0008] "Feature extraction" is the process of identifying and extracting important features such as color, shape, texture, and text from an image or slide.

[0009] "Vectorization" refers to converting data extracted as features into a numerical vector and making it into a format that can be calculated.

[0010] A "search query" is text information that a user enters when conducting a search.

[0011] "Receive as text" means that the system receives the character string data of the search query entered by the user.

[0012] "Similarity calculation" refers to a calculation that measures the similarity between two vectors, using techniques such as cosine similarity or Euclidean distance.

[0013] "Listing" means extracting and arranging related images and slides based on the results of the similarity calculation.

[0014] "Natural language processing technology" is a general term for technology that enables computers to understand and process human language, and is used for text analysis and vectorization.

[0015] A "deep learning model" is a machine learning model that uses a multi-layer neural network, and is a technology that demonstrates particularly high performance in extracting features from images and text. [Brief explanation of the drawings]

[0016] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11]FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0017] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0018] First, the terms used in the following description will be explained.

[0019] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple 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), and an APU (Accelerated Processing Unit).

[0020] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0021] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0022] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0023] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0024] [First embodiment]

[0025] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0026] 1, a 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.

[0027] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0028] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0029] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the 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.

[0030] 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 of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0031] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0032] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0033] 2, in the data processing device 12, a specific process 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" according to the technology of the present 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 process 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.

[0034] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0035] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the 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 process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0036] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0037] This invention relates to a system for efficiently searching for images and slides from past documents. This system includes a database for managing past documents, a user interface for inputting search queries, an algorithm for extracting and vectorizing features, and a function for calculating the similarity between the vectors of the documents and the search query.

[0038] System configuration

[0039] 1. Terminal

[0040] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0041] The terminal is provided with an input interface for the user to input a search query.

[0042] 2. Server

[0043] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0044] The server receives a search query from a user and vectorizes the query.

[0045] The server calculates the similarity between the vectorized query and the vectors of images or slides stored in the database.

[0046] Based on the results of the similarity calculation, the materials are listed in descending order of similarity.

[0047] 3. Database

[0048] The database stores vectorized data of past images and slides.

[0049] 4. Algorithm

[0050] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used to analyze and vectorize text queries.

[0051] Program processing

[0052] 1. Image / slide acquisition and feature extraction

[0053] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0054] The server extracts features from these images and slides, analyzing color, shape, texture, and text in the case of diagram images.

[0055] 2. Vectorization

[0056] The server converts the extracted features into vector format, for example, by using a deep learning model to convert image data into a multidimensional vector.

[0057] The extracted vector data is stored in a database.

[0058] 3. Entering a search query and vectorizing it

[0059] The user uses the input interface of the device to enter specific text describing the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0060] The server receives this text query and vectorizes it using natural language processing techniques, such as the BERT model, to convert the text into multidimensional vectors.

[0061] 4. Similarity calculation

[0062] The server calculates the similarity between the vectorized query and the vectors in the database. It measures the distance using cosine similarity or Euclidean distance, and calculates the similarity.

[0063] 5. Displaying the results

[0064] The server lists images and slides in order of similarity and sends the results to the device.

[0065] The terminal displays the results to the user, who can then check and select the images and slides they need.

[0066] Specific examples

[0067] For example, if a user is creating new planning materials on the theme of "future cities," they may want to reuse an "illustration of a future city" that was included in a previous document. The user enters text that describes the atmosphere of the "future city" into the search interface on their device. The server converts this text into vectors and calculates the similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can then select the "illustration of a future city" from the list and reuse it in the new document. This process allows users to find the materials they need quickly and efficiently.

[0068] The processing flow will be explained below.

[0069] Step 1:

[0070] The device retrieves images and slides from past documents, and loads multiple files containing graphs showing a company's growth prospects and illustrations of future cities.

[0071] Step 2:

[0072] The server extracts the features of the acquired images and slides. Using a deep learning model (e.g., Convolutional Neural Network, CNN), the server analyzes and extracts important features such as the color, shape, and texture of the images, as well as text information within the slides.

[0073] Step 3:

[0074] The server vectorizes the extracted features. The features are expressed as multidimensional vectors, and vectorized data is generated. This vector data is stored in a database.

[0075] Step 4:

[0076] A user uses a search interface on a device to enter a text query that specifies the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0077] Step 5:

[0078] The server analyzes and vectorizes the text query received from the user using natural language processing techniques. Specifically, it uses text encoding techniques such as BERT and Word2Vec to convert the text query into a multidimensional vector.

[0079] Step 6:

[0080] The server calculates the similarity between the vectorized query and the vector data of past documents in the database. It measures the distance between each vector using methods such as cosine similarity and Euclidean distance to calculate the similarity.

[0081] Step 7:

[0082] The server lists images and slides in descending order of similarity, generating a list of documents with the highest similarity first, and compiling this list as search results.

[0083] Step 8:

[0084] The server generates a search result list and sends it to the device, containing the images and slides that best match the user's search query.

[0085] Step 9:

[0086] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0087] In this way, even if a user cannot remember specific keywords, they can efficiently search for and reuse the necessary materials based on the atmosphere of the images and slides.

[0088] Example 1

[0089] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0090] Efficiently searching and retrieving necessary images and slides from past documents is a time-consuming and labor-intensive process. It is especially difficult to quickly find the desired content when a large amount of documents are stored in a database. Furthermore, accurate interpretation of user search queries and provision of optimal documents requires advanced technology.

[0091] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0092] In this invention, the server includes a means for uploading past materials specified by a user, a means for using a deep learning model to extract features of the uploaded materials, and a means for analyzing and vectorizing search queries using a natural language processing model, thereby enabling efficient search for necessary images and slides from past materials and providing the necessary materials with high accuracy.

[0093] "Images and slides" are visual content stored as past documents, and refer to graphics and presentation materials that provide concrete visual information.

[0094] "Features" are data that numerically represent specific properties or patterns extracted from digital data such as images or slides.

[0095] "Vectorization" is the process of representing features as vectors in a multidimensional space, which makes it easier to calculate similarities between data.

[0096] A "search query" is a text input used by a user to search for specific information, thereby providing guidance to the system in locating appropriate materials.

[0097] "Natural language processing technology" refers to a computational technology for analyzing and understanding human language, and in particular to the technology for semantically processing input text.

[0098] A "deep learning model" is a type of machine learning model that uses a multi-layer neural network to automatically learn complex patterns and features from data.

[0099] "Cosine similarity" is a method for calculating the similarity between two vectors, and uses the cosine value of the angle between the vectors.

[0100] "Euclidean distance" is a method for calculating the straight-line distance between two points (vectors). The smaller the distance, the more similar the two points (vectors) are considered to be.

[0101] A "database" is a system for efficiently storing, managing, and searching structured data.

[0102] This invention is a system that efficiently searches for images and slides from past documents and provides them to users. This system is composed of terminals, servers, databases, deep learning models, and natural language processing models.

[0103] First, the device retrieves images and slides from past documents, including hardware such as online storage and local file systems. The user can then specify the documents using the device's input interface and import them into the system.

[0104] The acquired images and slides are then processed on a server where features are extracted. The server uses a deep learning library (e.g., TENSORFLOW (registered trademark) or PyTorch) to analyze image features such as color, shape, texture, and text. These features are represented as multidimensional vectors.

[0105] The server then uses a pre-trained deep learning model (e.g., InceptionV3 or ResNet) to convert these extracted features into vector format, which is then stored in a database (e.g., MongoDB or MySQL®).

[0106] When a user enters a search query, they use the input interface of their device. For example, they enter specific text such as "illustration of a futuristic city" and click the search button. This text query is then vectorized by the server using a natural language processing model (e.g., BERT or Word2Vec).

[0107] The vectorized search query is used to calculate the similarity between the vector data of images and slides in the database. The server uses cosine similarity or Euclidean distance for this similarity calculation. Based on the calculated similarity, materials with high similarity are listed.

[0108] Finally, the server sends a list of materials with high similarity to the terminal, which displays it to the user. The user can then select and reuse the images and slides they need from the displayed materials. This system allows users to quickly and efficiently find the materials they need.

[0109] For example, consider a user creating new project materials on the theme of "future cities." The user enters "illustration of a future city" into the search interface on their device and clicks the search button. The server vectorizes this text and calculates its similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can select the "illustration of a future city" from the list and reuse it in the new material. This process allows users to quickly and efficiently find the materials they need.

[0110] Prompt Sentence Examples

[0111] "I want to create a planning document for a future city. Please search past documents for images and slides on the theme of future cities."

[0112] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0113] Step 1: Acquire images and slides

[0114] The device retrieves images and slides from past documents. The user uses a device such as a PC or smartphone to select a specific file from online storage or the local file system and clicks the upload button. In this step, the user can upload a "graph showing a company's growth prospects" or "illustration of a future city," and these files are sent to the system. The input data are the image or slide files, and the output data is a notification that the upload has been completed to the server.

[0115] Step 2: Feature extraction

[0116] The server analyzes images and slides received from the device and extracts features. Deep learning libraries (e.g., TensorFlow and PyTorch) are used for extraction. The image's color distribution, shape, texture, and included text are automatically analyzed. The input data are the uploaded image or slide files, and the output data is a dataset containing the extracted features.

[0117] Step 3: Vectorization

[0118] The server converts the features into multidimensional vectors. A pre-trained deep learning model (e.g., InceptionV3 or ResNet) is used to map the extracted features into a high-dimensional vector space. The input data are the features extracted in step 2, and the output data is data represented as multidimensional vectors.

[0119] Step 4: Saving to the Database

[0120] The server stores the vectorized data in a database (e.g., MongoDB or MySQL). At this time, metadata (e.g., title, date, category, etc.) of the corresponding image or slide is also stored for each vector. The input data is data and metadata expressed as multidimensional vectors, and the output data is vector data stored in the database.

[0121] Step 5: Enter a search query

[0122] The user enters a search query using the search interface on the device. For example, they enter specific keywords such as "illustration of a futuristic city" or "image showing a bright future," and clicks the search button. The input data is the text of the search query, and the output data is a notification that the query has been sent to the server.

[0123] Step 6: Vectorizing the query

[0124] The server vectorizes the search query received from the user using a natural language processing model (e.g., BERT or Word2Vec). It captures the meaning of the query and represents the text as a multidimensional vector. The input data is the search query text, and the output data is the vectorized query.

[0125] Step 7: Similarity calculation

[0126] The server calculates the similarity between the vectorized query and the vector data in the database using algorithms such as cosine similarity or Euclidean distance. The input data are the vectorized query and the vector data in the database, and the output data is the similarity score.

[0127] Step 8: Viewing the results

[0128] The server lists images and slides in descending order of similarity based on the similarity score and sends the list to the terminal. The terminal displays the results to the user, who can then select the images or slides they need from the displayed list. The input data is the similarity list generated by the server, and the output data is the results displayed on the user's terminal.

[0129] (Application example 1)

[0130] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0131] In today's brick-and-mortar stores, it is difficult for consumers to efficiently search for products based on their preferences and past purchase history. It is also difficult for them to instantly understand the vast amount of product information available in stores and make appropriate selections. Therefore, there is a need for a method that allows consumers to find products more quickly and accurately.

[0132] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0133] In this invention, the server includes means for acquiring images and slides from past materials, means for extracting features from the acquired images and slides and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for calculating the similarity between the vectorized search query and the vectors of the past materials and listing materials with high similarity, means for displaying materials with high similarity, means for searching for retail products based on images, means for vectorizing an image taken as the search query and calculating the similarity with product vectors in a database, and means for listing and displaying products with high similarity. This allows a user to instantly find similar products based on images of products they found in a store.

[0134] "Past materials" refers to previously created images, slides, and document data.

[0135] "Features of images and slides" refer to important elements that make up an image, such as color, shape, texture, and text, extracted from the image or slide.

[0136] "Vectorization" is the process of converting features into points in a multidimensional space.

[0137] A "search query" is the text or image information that a user enters into the system.

[0138] "Similarity" is an index that measures the similarity between two vectors, and there are calculation methods such as cosine similarity and Euclidean distance.

[0139] A "database" is a digital warehouse that stores and manages vectorized data extracted from past documents.

[0140] "Natural language processing technology" is a technology that understands and analyzes text data and converts it into a format that can be processed by machines.

[0141] A "deep learning model" is an artificial intelligence technology that automatically learns features using large amounts of data and makes highly accurate predictions and classifications.

[0142] "Retail" is a general term that refers to retail trade and brick-and-mortar sales.

[0143] A "means for calculating similarity" refers to a method or technique for comparing a vectorized query with a vector in a database and measuring the similarity between them.

[0144] This invention relates to a system for improving the shopping experience in brick-and-mortar stores. The system aims to efficiently search for and present related and recommended products to users based on their past purchase history and images and descriptions of products seen in stores.

[0145] The system uses a smartphone as hardware and databases (e.g., MongoDB), deep learning models (e.g., CNN using TensorFlow), and natural language processing models (e.g., BERT) as software. It uses Node.js and Express on the server side and React Native on the front end.

[0146] A user uses the camera function of their smartphone to take a picture of a product they see in a store. This image is sent to the server via the device. The server extracts features from the received image and vectorizes them using a deep learning model. These features include the product's shape, color, texture, etc.

[0147] The server stores the vectorized feature data in a database. Furthermore, the server receives search queries (e.g., "red dress" or "sports shoes") entered by users as text and vectorizes them using natural language processing techniques. The BERT model is used for this vectorization.

[0148] The server calculates the similarity between the vectorized search query and the product vectors in the database. This similarity calculation uses cosine similarity or Euclidean distance. The server then lists the products in descending order of similarity and sends the results to the smartphone.

[0149] The smartphone displays products sorted by similarity, and users can select the product they are interested in and check its details. This allows users to instantly find similar products based on the image of the product they found in the store.

[0150] For example, if a user is looking for a "red dress" in a store, they can take a photo of the product with their smartphone and enter the keyword "red dress." The server analyzes the features based on the image and keyword, and vectorizes them using a deep learning model and natural language processing technology. The server then calculates the similarity with products in the database and displays the results on the user's smartphone.

[0151] Examples of prompts include:

[0152] I'm looking for a "red dress." Show me products similar to this in-store image.

[0153] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0154] Step 1:

[0155] A user takes a photo of a product in a store using the camera on their smartphone.

[0156] Input: Image taken inside the store

[0157] Output: Captured image data

[0158] Image data is sent to the server, which receives the data and begins processing it.

[0159] Step 2:

[0160] The server extracts features from the received image data.

[0161] Input: Photographed image data

[0162] Output: Feature vector

[0163] The server uses a deep learning model (e.g., CNN using TensorFlow) to extract features such as color, shape, and texture from the image.

[0164] Step 3:

[0165] The server vectorizes the extracted features.

[0166] Input: extracted features

[0167] Output: A vector of features

[0168] The server converts these features into multidimensional vectors, which are then stored in a MongoDB database.

[0169] Step 4:

[0170] A user inputs a search query through an input interface of a smartphone.

[0171] Input: Search query (e.g. "red dress")

[0172] Output: Query text

[0173] The terminal transmits this text data to the server.

[0174] Step 5:

[0175] The text data of the search query received by the server is vectorized.

[0176] Input: Text data of the search query

[0177] Output: A vector of text

[0178] The server analyzes this text using natural language processing techniques (e.g., the BERT model) and vectorizes it.

[0179] Step 6:

[0180] The server calculates the similarity between the vectorized search query and the product vectors in the database.

[0181] Input: Search query vector and product feature vector data

[0182] Output: Similarity score

[0183] The server calculates the similarity between the query vector and the product vector using cosine similarity or Euclidean distance.

[0184] Step 7:

[0185] The server lists products in order of similarity and sends the results to the smartphone.

[0186] Input: Similarity score

[0187] Output: List of similar products

[0188] The server lists products with high similarity and sends the results to the terminal.

[0189] Step 8:

[0190] The terminal displays the search results to the user.

[0191] Input: List of similar products

[0192] Output: Displayed product list

[0193] Users can view search results on their smartphones, select products that interest them, and read more details.

[0194] This system allows users to instantly find similar products based on images of products they find in stores.

[0195] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0196] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, the emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0197] System configuration

[0198] 1. Terminal

[0199] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0200] The user interface provides a means for users to enter search queries and also includes an interface for accepting facial recognition and voice input from users.

[0201] 2. Server

[0202] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0203] The search query received from the user interface is received as text, and the user's emotion recognized from the emotion engine is also received.

[0204] The server vectorizes the search query and corrects it based on sentiment data.

[0205] The similarity between the vectorized query and the vectors of past materials is calculated, and materials with high similarity are listed.

[0206] 3. Database

[0207] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[0208] 4. Algorithm

[0209] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for analyzing text queries.

[0210] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0211] Program processing

[0212] 1. Image / slide acquisition and feature extraction

[0213] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0214] The server extracts features from these images and slides. For images, it analyzes and extracts color, shape, texture, and text information from the slides.

[0215] 2. Vectorization

[0216] The server converts the extracted features into vector format, which allows the image data to be represented as a multidimensional vector.

[0217] These vector data are stored in a database.

[0218] 3. Entering a search query and vectorizing it

[0219] A user uses a device to input a text query describing the image or slide atmosphere they want to search for. For example, they might input the text, "I want an image that depicts a bright future."

[0220] The device sends the user's facial expression and voice data to the emotion engine to recognize the user's emotions. For example, if a user searches for "bright future" with a smile, the emotion engine will recognize this as a "positive emotion."

[0221] The server receives the search query and sentiment data, vectorizes the text query, and corrects it based on sentiment.

[0222] 4. Similarity calculation

[0223] The server calculates the similarity between the vectorized query and the vectors in the database, using methods such as cosine similarity or Euclidean distance.

[0224] 5. Displaying the results

[0225] The server lists images and slides in order of similarity and sends the results to the device.

[0226] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0227] Specific examples

[0228] For example, if a user is creating a new project document with the theme of "future cities," they may want to reuse an illustration of a future city from a previous document. The user enters text that describes the atmosphere of a "future city" into the device's search interface and intentionally smiles. The device recognizes the facial expression and sends it to the server as positive emotion data. The server receives the text query and emotion data and generates appropriate search results based on them. This allows the illustration the user is looking for to be efficiently reused from previous documents. This process allows users to find the materials they need quickly and efficiently.

[0229] The processing flow will be explained below.

[0230] Step 1:

[0231] The terminal retrieves images and slides from past documents. Specifically, it accesses the database or file system within the system and reads images and slides contained in planning documents and proposal documents.

[0232] Step 2:

[0233] The server extracts features from the acquired images and slides. The server uses a deep learning model (e.g., Convolutional Neural Network, CNN) to analyze the color, shape, and texture of the images, as well as the text information in the slides, and extracts features.

[0234] Step 3:

[0235] The server vectorizes the extracted features, expresses the features as multidimensional vectors, and generates vectorized data. These vector data are stored in a database.

[0236] Step 4:

[0237] A user uses the device's search interface to enter a text query that specifies the image or slide mood they are looking for, for example, "I want images that depict a bright future."

[0238] Step 5:

[0239] The emotion engine receives user input and analyzes the user's facial expression and voice data to recognize emotions. Specifically, it captures the user's facial expressions through a camera and identifies emotions using a facial expression analysis algorithm. It also captures voice data through a microphone and analyzes emotions using a voice emotion recognition algorithm.

[0240] Step 6:

[0241] The server receives the user's text query and the recognized emotion data from the emotion engine, vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec), and corrects the search query based on the recognized emotion.

[0242] Step 7:

[0243] The server calculates the similarity between the vectorized query and the data vectors in the database, using methods such as cosine similarity or Euclidean distance to measure the distance between each vector and calculate the similarity.

[0244] Step 8:

[0245] The server lists images and slides in order of similarity, compiles the results as search results, generates a search result list, and sends the results to the device.

[0246] Step 9:

[0247] The terminal displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0248] Specific examples

[0249] For example, when creating new planning materials with the theme of "future cities," a user might want to reuse an "illustration of a future city." The user enters text describing the atmosphere of the "future city" into the device's search interface and then captures their own smile with the camera. The emotion engine recognizes the user's smile and generates emotional data representing a "positive emotion." The server receives the text query and the emotional data, and corrects the search query based on both to generate optimal search results. As a result, the "illustration of a future city" desired by the user is efficiently listed and can be reused in creating new planning materials. This process, with the help of the emotion engine, enables searches that more accurately reflect the user's intent.

[0250] Example 2

[0251] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0252] Conventional document search systems not only have difficulty efficiently finding the necessary images and slides from a large number of digital files, but also fail to consider the user's emotions, resulting in search results that often deviate from the user's intentions. Furthermore, feature extraction and search query vectorization rely too heavily on user input, resulting in poor search result accuracy.

[0253] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0254] In this invention, the server includes means for acquiring digital files from past materials, means for extracting features from the acquired digital files and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for recognizing the user's emotions and generating emotion data, means for correcting the vectorized search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past materials and listing materials with high similarity, and means for displaying materials with high similarity. This makes it possible to provide highly accurate search results while taking user emotions into consideration.

[0255] "Past materials" are digital files that were created or collected in the past and registered in the system, and primarily refer to images and slides.

[0256] "Digital files" refers to data stored and exchanged electronically, and generally includes image files and slide files.

[0257] A "feature" is an attribute value or pattern extracted from a digital file, and is information that allows the characteristics of the digital file to be quantified or vectorized.

[0258] "Vectorization" is a technique that represents digital files or text queries as multidimensional numerical vectors, making them easier to process on a computer.

[0259] "Search query" refers to a text input by a user that indicates to the system what they wish to search for.

[0260] "Emotion data" refers to data that expresses the emotional state recognized from the user's facial expressions and voice as numerical values ​​or categories.

[0261] "Similarity" is an index that indicates the closeness or degree of agreement between two vectors (a search query and a digital file's features), and is calculated using methods such as cosine similarity or Euclidean distance.

[0262] "Listing" refers to displaying a list of digital files obtained as search results in order of similarity.

[0263] This invention is a system for efficiently searching for images and slides from past documents, equipped with a function for recognizing user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion recognition engine, and a function for calculating the similarity between vectors. The hardware and software used are described below.

[0264] Hardware and Software Configuration

[0265] 1. Terminal

[0266] Devices used as terminals include PCs, tablets, smartphones, etc.

[0267] The device provides a user interface (UI) for users to enter search queries, which includes not only text input but also facial expression recognition and voice input functions.

[0268] 2. Server

[0269] The server is the main component that processes the data sent from the terminals.

[0270] Deep learning models (e.g., Convolutional Neural Network, CNN) are used to extract features from images and slides.

[0271] Natural language processing techniques (e.g., BERT, Word2Vec) are used to parse and vectorize text queries.

[0272] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0273] 3. Database

[0274] The features of past images and slides are stored in vector format, and the database also includes an index that allows fast and efficient search of vector data.

[0275] Specific operation of the system

[0276] Suppose a user searches for "an illustration that expresses the atmosphere of a futuristic city." The user enters a text query through the device's search interface and intentionally smiles. The device recognizes this facial expression in real time and transmits it to the server as positive emotion data.

[0277] The server vectorizes the text query and the received emotion data, and then modifies the query based on the emotion data. The server then compares the vectors with those of past digital files to identify those with high similarity, using, for example, cosine similarity or Euclidean distance, to generate optimal search results.

[0278] Finally, the server sends the search results to the terminal, which displays them to the user. The user can select images or slides that suit their purpose from the search results and reuse them in new materials. This allows users to quickly and accurately find the materials they need.

[0279] Example prompt sentence:

[0280] Search for "I want an illustration that conveys the atmosphere of a futuristic city."

[0281] Please implement a system that allows users to search for "future cities" with a smile.

[0282] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0283] Step 1: User enters a search query

[0284] The user uses the device to input the desired image or slide content as text into the search interface. For example, they might input "I'm looking for an illustration that evokes the atmosphere of a futuristic city." At this point, the search query text is generated as input data. Once the user has finished entering the query, the device passes this input data to the next processing step.

[0285] Step 2: Recognizing user emotion data by the device

[0286] As the user enters a text query, the device's camera and microphone capture the user's facial expressions and voice. The device then uses an emotion recognition engine to analyze this data and recognize the user's emotion. The emotion data generated is then sent to the next processing step.

[0287] Step 3: Obtaining past documents via terminal

[0288] The device retrieves previously created materials (digital files) from the company's cloud storage or internal server. The input here is past materials retrieved according to conditions specified by the user. The retrieved past materials (images and slides) are sent to the next processing step.

[0289] Step 4: Feature extraction and vectorization on the server

[0290] The server extracts features from images and slides of past documents received from the terminal. A deep learning model (e.g., CNN) is used to extract features. At this time, the input data is past documents that meet the user's requirements, and the output data is the features extracted from these documents. The features are converted into vector format and stored in a database.

[0291] Step 5: Vectorizing search queries by the server

[0292] The server vectorizes the text query entered by the user using natural language processing techniques (e.g., BERT, Word2Vec). Here, the input is the text query and the output is the vectorized query. This vectorized query is sent to the next processing step.

[0293] Step 6: Server-based query refinement based on emotion data

[0294] The server uses the sentiment data to correct the vectorized query. For example, if there is positive sentiment data, it adds a positive weight to the corresponding search query vector. Here, the input is the vectorized query and sentiment data, and the output is the corrected vector query.

[0295] Step 7: Similarity calculation by the server

[0296] The server calculates the similarity between the corrected vector query and the document vectors in the database. The algorithms used include cosine similarity and Euclidean distance. The input data are the corrected query vector and the document vector, and the output is a list of documents with high similarity.

[0297] Step 8: Server lists and sends search results

[0298] As a result of the similarity calculation, the server lists the materials in descending order of similarity and sends it to the terminal. The input here is the result of the similarity calculation, and the output is a list of search results. The list of listed materials is sent to the terminal.

[0299] Step 9: Device displays search results and user selection

[0300] The terminal displays the received search results on the user interface. The user can select images or slides that suit their purpose from this list. The input here is the list of search results, and the output is the materials selected by the user. The user can reuse the selected materials in new project materials to create.

[0301] (Application example 2)

[0302] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0303] When searching for images or slides from past documents, it is difficult to obtain appropriate results that reflect the user's emotions. In particular, if search results are generated only by simple text matching or vector comparison without considering the user's emotions when searching, the optimal results may not be provided according to the user's expectations.

[0304] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0305] In this invention, the server includes means for extracting and vectorizing acquired images and slides as features, means for vectorizing search queries entered by users, means for recognizing user emotions and correcting the search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past documents and listing documents with high similarity, and means for receiving user emotions as feedback and reflecting them in the next search, thereby making it possible to provide optimal search results that reflect the user's emotions.

[0306] - "Historical materials" refers to data such as digital documents, images, and slides that have been collected and stored to date.

[0307] "Image" refers to any visual data displayed or stored by a digital device.

[0308] "Slides" refers to individual visual pages used in presentations and reports.

[0309] "Means of acquisition" refers to a method or device for capturing data of historical materials from a designated location.

[0310] "Means for extracting features" refers to a method or device that analyzes the visual and content elements of an image or slide and converts them into numerical data.

[0311] "Vectorization means" refers to a method or device that formalizes extracted features as multidimensional vectors.

[0312] "Search query" refers to a string of characters or spoken instructions entered by a user to indicate the information or material they are seeking.

[0313] "Means for recognizing emotions" refers to technology or devices that determine a user's psychological state at that time from facial expressions and voice data.

[0314] "Means for correcting based on emotional data" refers to technologies or devices that adjust or transform search queries taking into account the emotional state of the user.

[0315] The term "means for calculating similarity" refers to an algorithm or device for numerically evaluating the similarity between vector-format data.

[0316] The "listing means" refers to a method or device for preferentially extracting data with high similarity and displaying it in a list format.

[0317] "Displaying means" refers to a method or device for visually presenting search results to a user.

[0318] "Means for receiving feedback" refers to a method or device for incorporating user evaluations and opinions into the system.

[0319] A "deep learning model" refers to an algorithm that uses an advanced neural network to extract and learn features.

[0320] "Natural language processing technology" refers to all technologies for analyzing and understanding natural language texts that humans use on a daily basis.

[0321] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. The system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0322] System configuration

[0323] 1. Terminal

[0324] The terminal retrieves previously created planning and proposal materials and sends them to the system. The user interface provides a means for users to input search queries. It also includes an interface that recognizes the user's facial expressions and receives voice input. Specifically, a smartphone or smart glasses can function as the terminal.

[0325] 2. Server

[0326] The server extracts features from images and slides obtained from the device and vectorizes them. A deep learning model (e.g., CNN) is used to extract features, and natural language processing technology (e.g., BERT, Word2Vec) is used to analyze text queries. The server receives the search query as text from the user interface, as well as the user's emotions recognized by the emotion engine. The server vectorizes the search query and corrects it based on the emotion data. The server calculates the similarity between the vectorized query and the vectors of past materials and lists materials with high similarity.

[0327] 3. Database

[0328] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[0329] 4. Algorithm

[0330] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for text query analysis. The emotion engine uses technologies to recognize user facial expressions and identify emotions from voice (e.g., facial expression analysis, voice emotion recognition).

[0331] Program processing

[0332] The server uses a deep learning model to extract features from images and slides sent from the device and vectorizes them. The device receives the search query entered by the user through the user interface as text and recognizes the user's emotions using an emotion engine. The server receives this text query and emotion data, vectorizes the query using natural language processing technology, and corrects it based on the emotion data. The server then calculates the similarity between the vectorized query and the vectors of past documents and lists documents with high similarity. Finally, the search results are displayed to the user via the device. The user can select images or slides that match their purpose from the displayed results and reuse them in creating documents.

[0333] Specific examples

[0334] For example, imagine a user is searching for "new sneakers" in a physical store. The user launches a smartphone app, enters "comfortable sneakers" in the search field, and intentionally smiles. The app recognizes this facial expression and sends it to the server as a positive emotion.

[0335] Example prompt (input to generative AI model):

[0336] A user is searching for "comfortable sneakers." The emotion engine recognizes the user's smile and confirms that the emotion is positive. Find the most suitable sneakers based on past purchase history and product reviews. Also, display related product recommendations in the search results.

[0337] This process allows users to quickly find the product that best suits their emotions, improving the shopping experience in physical stores. Specifically, it lists comfortable sneakers and related products that users have previously purchased, and allows users to try them on and provide feedback to further improve the service.

[0338] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0339] Step 1:

[0340] The terminal retrieves images and slides from past documents and sends this data to the system. Specifically, it retrieves previously created planning and proposal documents from a database. The input is past document data, and the output is data sent to the server.

[0341] Step 2:

[0342] The server uses a deep learning model (e.g., CNN) to extract features from acquired images and slides and vectorize them. Specifically, it analyzes and extracts image color, shape, texture, and text information from slides. The input is image or slide data, and the output is feature data in vector format.

[0343] Step 3:

[0344] The user uses the device to input the mood of the image or slide they want to search for as a text query. At this time, the device also sends the user's facial expressions and voice input to the emotion engine to recognize emotions. The input is the user's text query and emotion data, and the output is data sent from the device to the server.

[0345] Step 4:

[0346] The server analyzes and vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec). It then corrects the vector based on emotion data from the emotion engine. The input is the text query and emotion data, and the output is the corrected vector-formatted search query.

[0347] Step 5:

[0348] The server calculates the similarity between the vectorized search query and the vectors of past documents using methods such as cosine similarity or Euclidean distance. The input is the corrected search query vector and the vectors of past documents, and the output is a list of documents with high similarity.

[0349] Step 6:

[0350] The server lists the search results in order of similarity and sends them to the terminal. The input is a list of similarities between documents, and the output is the transmission of the search results to the terminal.

[0351] Step 7:

[0352] The terminal displays the search results to the user. The user selects images and slides that fit the purpose from the displayed list and reuses them in creating materials. The input is the search results sent from the server, and the output is a visual presentation to the user.

[0353] Step 8:

[0354] The user inputs feedback such as satisfaction level for the selected result, and the terminal transmits this feedback to the server. The input is the user's feedback data, and the output is data transmission to the server.

[0355] Step 9:

[0356] The server reflects the received feedback in the emotion engine to improve the accuracy of the next search results. The input is the feedback data, and the output is an improved emotion model.

[0357] 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 a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the 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.

[0358] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (registered trademark) (Internet search engine).<URL: https: / / openai.com / blog / chatgpt> ), Gemini (registered trademark) (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0359] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0360] [Second embodiment]

[0361] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0362] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0363] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0364] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0365] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0366] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0367] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0368] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0369] The specific processing program 56 is an example of a "program" according to the technology of the present 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.

[0370] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0371] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. 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 process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0372] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. 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."

[0373] This invention relates to a system for efficiently searching for images and slides from past documents. This system includes a database for managing past documents, a user interface for inputting search queries, an algorithm for extracting and vectorizing features, and a function for calculating the similarity between the vectors of the documents and the search query.

[0374] System configuration

[0375] 1. Terminal

[0376] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0377] The terminal is provided with an input interface for the user to input a search query.

[0378] 2. Server

[0379] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0380] The server receives a search query from a user and vectorizes the query.

[0381] The server calculates the similarity between the vectorized query and the vectors of images or slides stored in the database.

[0382] Based on the results of the similarity calculation, the materials are listed in descending order of similarity.

[0383] 3. Database

[0384] The database stores vectorized data of past images and slides.

[0385] 4. Algorithm

[0386] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used to analyze and vectorize text queries.

[0387] Program processing

[0388] 1. Image / slide acquisition and feature extraction

[0389] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0390] The server extracts features from these images and slides, analyzing color, shape, texture, and text in the case of diagram images.

[0391] 2. Vectorization

[0392] The server converts the extracted features into vector format, for example, by using a deep learning model to convert image data into a multidimensional vector.

[0393] The extracted vector data is stored in a database.

[0394] 3. Entering a search query and vectorizing it

[0395] The user uses the input interface of the device to enter specific text describing the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0396] The server receives this text query and vectorizes it using natural language processing techniques, such as the BERT model, to convert the text into multidimensional vectors.

[0397] 4. Similarity calculation

[0398] The server calculates the similarity between the vectorized query and the vectors in the database. It measures the distance using cosine similarity or Euclidean distance, and calculates the similarity.

[0399] 5. Displaying the results

[0400] The server lists images and slides in order of similarity and sends the results to the device.

[0401] The terminal displays the results to the user, who can then check and select the images and slides they need.

[0402] Specific examples

[0403] For example, if a user is creating new planning materials on the theme of "future cities," they may want to reuse an "illustration of a future city" that was included in a previous document. The user enters text that describes the atmosphere of the "future city" into the search interface on their device. The server converts this text into vectors and calculates the similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can then select the "illustration of a future city" from the list and reuse it in the new document. This process allows users to find the materials they need quickly and efficiently.

[0404] The processing flow will be explained below.

[0405] Step 1:

[0406] The device retrieves images and slides from past documents, and loads multiple files containing graphs showing a company's growth prospects and illustrations of future cities.

[0407] Step 2:

[0408] The server extracts the features of the acquired images and slides. Using a deep learning model (e.g., Convolutional Neural Network, CNN), the server analyzes and extracts important features such as the color, shape, and texture of the images, as well as text information within the slides.

[0409] Step 3:

[0410] The server vectorizes the extracted features. The features are expressed as multidimensional vectors, and vectorized data is generated. This vector data is stored in a database.

[0411] Step 4:

[0412] A user uses a search interface on a device to enter a text query that specifies the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0413] Step 5:

[0414] The server analyzes and vectorizes the text query received from the user using natural language processing techniques. Specifically, it uses text encoding techniques such as BERT and Word2Vec to convert the text query into a multidimensional vector.

[0415] Step 6:

[0416] The server calculates the similarity between the vectorized query and the vector data of past documents in the database. It measures the distance between each vector using methods such as cosine similarity and Euclidean distance to calculate the similarity.

[0417] Step 7:

[0418] The server lists images and slides in descending order of similarity, generating a list of documents with the highest similarity first, and compiling this list as search results.

[0419] Step 8:

[0420] The server generates a search result list and sends it to the device, containing the images and slides that best match the user's search query.

[0421] Step 9:

[0422] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0423] In this way, even if a user cannot remember specific keywords, they can efficiently search for and reuse the necessary materials based on the atmosphere of the images and slides.

[0424] Example 1

[0425] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0426] Efficiently searching and retrieving necessary images and slides from past documents is a time-consuming and labor-intensive process. It is especially difficult to quickly find the desired content when a large amount of documents are stored in a database. Furthermore, accurate interpretation of user search queries and provision of optimal documents requires advanced technology.

[0427] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0428] In this invention, the server includes a means for uploading past materials specified by a user, a means for using a deep learning model to extract features of the uploaded materials, and a means for analyzing and vectorizing search queries using a natural language processing model, thereby enabling efficient search for necessary images and slides from past materials and providing the necessary materials with high accuracy.

[0429] "Images and slides" are visual content stored as past documents, and refer to graphics and presentation materials that provide concrete visual information.

[0430] "Features" are data that numerically represent specific properties or patterns extracted from digital data such as images or slides.

[0431] "Vectorization" is the process of representing features as vectors in a multidimensional space, which makes it easier to calculate similarities between data.

[0432] A "search query" is a text input used by a user to search for specific information, thereby providing guidance to the system in locating appropriate materials.

[0433] "Natural language processing technology" refers to a computational technology for analyzing and understanding human language, and in particular to the technology for semantically processing input text.

[0434] A "deep learning model" is a type of machine learning model that uses a multi-layer neural network to automatically learn complex patterns and features from data.

[0435] "Cosine similarity" is a method for calculating the similarity between two vectors, and uses the cosine value of the angle between the vectors.

[0436] "Euclidean distance" is a method for calculating the straight-line distance between two points (vectors). The smaller the distance, the more similar the two points (vectors) are considered to be.

[0437] A "database" is a system for efficiently storing, managing, and searching structured data.

[0438] This invention is a system that efficiently searches for images and slides from past documents and provides them to users. This system is composed of terminals, servers, databases, deep learning models, and natural language processing models.

[0439] First, the device retrieves images and slides from past documents, including hardware such as online storage and local file systems. The user can then specify the documents using the device's input interface and import them into the system.

[0440] The captured images and slides are then processed on a server where features are extracted. The server uses deep learning libraries (e.g., TensorFlow and PyTorch) to analyze image features such as color, shape, texture, and text. These features are represented as multidimensional vectors.

[0441] The server then uses a pre-trained deep learning model (e.g., InceptionV3 or ResNet) to convert these extracted features into vector format, which is then stored in a database (e.g., MongoDB or MySQL).

[0442] When a user enters a search query, they use the input interface of their device. For example, they enter specific text such as "illustration of a futuristic city" and click the search button. This text query is then vectorized by the server using a natural language processing model (e.g., BERT or Word2Vec).

[0443] The vectorized search query is used to calculate the similarity between the vector data of images and slides in the database. The server uses cosine similarity or Euclidean distance for this similarity calculation. Based on the calculated similarity, materials with high similarity are listed.

[0444] Finally, the server sends a list of materials with high similarity to the terminal, which displays it to the user. The user can then select and reuse the images and slides they need from the displayed materials. This system allows users to quickly and efficiently find the materials they need.

[0445] For example, consider a user creating new project materials on the theme of "future cities." The user enters "illustration of a future city" into the search interface on their device and clicks the search button. The server vectorizes this text and calculates its similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can select the "illustration of a future city" from the list and reuse it in the new material. This process allows users to quickly and efficiently find the materials they need.

[0446] Prompt Sentence Examples

[0447] "I want to create a planning document for a future city. Please search past documents for images and slides on the theme of future cities."

[0448] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0449] Step 1: Acquire images and slides

[0450] The device retrieves images and slides from past documents. The user uses a device such as a PC or smartphone to select a specific file from online storage or the local file system and clicks the upload button. In this step, the user can upload a "graph showing a company's growth prospects" or "illustration of a future city," and these files are sent to the system. The input data are the image or slide files, and the output data is a notification that the upload has been completed to the server.

[0451] Step 2: Feature extraction

[0452] The server analyzes images and slides received from the device and extracts features. Deep learning libraries (e.g., TensorFlow and PyTorch) are used for extraction. The image's color distribution, shape, texture, and included text are automatically analyzed. The input data are the uploaded image or slide files, and the output data is a dataset containing the extracted features.

[0453] Step 3: Vectorization

[0454] The server converts the features into multidimensional vectors. A pre-trained deep learning model (e.g., InceptionV3 or ResNet) is used to map the extracted features into a high-dimensional vector space. The input data are the features extracted in step 2, and the output data is data represented as multidimensional vectors.

[0455] Step 4: Saving to the Database

[0456] The server stores the vectorized data in a database (e.g., MongoDB or MySQL). At this time, metadata (e.g., title, date, category, etc.) of the corresponding image or slide is also stored for each vector. The input data is data and metadata expressed as multidimensional vectors, and the output data is vector data stored in the database.

[0457] Step 5: Enter a search query

[0458] The user enters a search query using the search interface on the device. For example, they enter specific keywords such as "illustration of a futuristic city" or "image showing a bright future," and clicks the search button. The input data is the text of the search query, and the output data is a notification that the query has been sent to the server.

[0459] Step 6: Vectorizing the query

[0460] The server vectorizes the search query received from the user using a natural language processing model (e.g., BERT or Word2Vec). It captures the meaning of the query and represents the text as a multidimensional vector. The input data is the search query text, and the output data is the vectorized query.

[0461] Step 7: Similarity calculation

[0462] The server calculates the similarity between the vectorized query and the vector data in the database using algorithms such as cosine similarity or Euclidean distance. The input data are the vectorized query and the vector data in the database, and the output data is the similarity score.

[0463] Step 8: Viewing the results

[0464] The server lists images and slides in descending order of similarity based on the similarity score and sends the list to the terminal. The terminal displays the results to the user, who can then select the images or slides they need from the displayed list. The input data is the similarity list generated by the server, and the output data is the results displayed on the user's terminal.

[0465] (Application example 1)

[0466] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0467] In today's brick-and-mortar stores, it is difficult for consumers to efficiently search for products based on their preferences and past purchase history. It is also difficult for them to instantly understand the vast amount of product information available in stores and make appropriate selections. Therefore, there is a need for a method that allows consumers to find products more quickly and accurately.

[0468] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0469] In this invention, the server includes means for acquiring images and slides from past materials, means for extracting features from the acquired images and slides and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for calculating the similarity between the vectorized search query and the vectors of the past materials and listing materials with high similarity, means for displaying materials with high similarity, means for searching for retail products based on images, means for vectorizing an image taken as the search query and calculating the similarity with product vectors in a database, and means for listing and displaying products with high similarity. This allows a user to instantly find similar products based on images of products they found in a store.

[0470] "Past materials" refers to previously created images, slides, and document data.

[0471] "Features of images and slides" refer to important elements that make up an image, such as color, shape, texture, and text, extracted from the image or slide.

[0472] "Vectorization" is the process of converting features into points in a multidimensional space.

[0473] A "search query" is the text or image information that a user enters into the system.

[0474] "Similarity" is an index that measures the similarity between two vectors, and there are calculation methods such as cosine similarity and Euclidean distance.

[0475] A "database" is a digital warehouse that stores and manages vectorized data extracted from past documents.

[0476] "Natural language processing technology" is a technology that understands and analyzes text data and converts it into a format that can be processed by machines.

[0477] A "deep learning model" is an artificial intelligence technology that automatically learns features using large amounts of data and makes highly accurate predictions and classifications.

[0478] "Retail" is a general term that refers to retail trade and brick-and-mortar sales.

[0479] A "means for calculating similarity" refers to a method or technique for comparing a vectorized query with a vector in a database and measuring the similarity between them.

[0480] This invention relates to a system for improving the shopping experience in brick-and-mortar stores. The system aims to efficiently search for and present related and recommended products to users based on their past purchase history and images and descriptions of products seen in stores.

[0481] The system uses a smartphone as hardware and databases (e.g., MongoDB), deep learning models (e.g., CNN using TensorFlow), and natural language processing models (e.g., BERT) as software. It uses Node.js and Express on the server side and React Native on the front end.

[0482] A user uses the camera function of their smartphone to take a picture of a product they see in a store. This image is sent to the server via the device. The server extracts features from the received image and vectorizes them using a deep learning model. These features include the product's shape, color, texture, etc.

[0483] The server stores the vectorized feature data in a database. Furthermore, the server receives search queries (e.g., "red dress" or "sports shoes") entered by users as text and vectorizes them using natural language processing techniques. The BERT model is used for this vectorization.

[0484] The server calculates the similarity between the vectorized search query and the product vectors in the database. This similarity calculation uses cosine similarity or Euclidean distance. The server then lists the products in descending order of similarity and sends the results to the smartphone.

[0485] The smartphone displays products sorted by similarity, and users can select the product they are interested in and check its details. This allows users to instantly find similar products based on the image of the product they found in the store.

[0486] For example, if a user is looking for a "red dress" in a store, they can take a photo of the product with their smartphone and enter the keyword "red dress." The server analyzes the features based on the image and keyword, and vectorizes them using a deep learning model and natural language processing technology. The server then calculates the similarity with products in the database and displays the results on the user's smartphone.

[0487] Examples of prompts include:

[0488] I'm looking for a "red dress." Show me products similar to this in-store image.

[0489] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0490] Step 1:

[0491] A user takes a photo of a product in a store using the camera on their smartphone.

[0492] Input: Image taken inside the store

[0493] Output: Captured image data

[0494] Image data is sent to the server, which receives the data and begins processing it.

[0495] Step 2:

[0496] The server extracts features from the received image data.

[0497] Input: Photographed image data

[0498] Output: Feature vector

[0499] The server uses a deep learning model (e.g., CNN using TensorFlow) to extract features such as color, shape, and texture from the image.

[0500] Step 3:

[0501] The server vectorizes the extracted features.

[0502] Input: extracted features

[0503] Output: A vector of features

[0504] The server converts these features into multidimensional vectors, which are then stored in a MongoDB database.

[0505] Step 4:

[0506] A user inputs a search query through an input interface of a smartphone.

[0507] Input: Search query (e.g. "red dress")

[0508] Output: Query text

[0509] The terminal transmits this text data to the server.

[0510] Step 5:

[0511] The text data of the search query received by the server is vectorized.

[0512] Input: Text data of the search query

[0513] Output: A vector of text

[0514] The server analyzes this text using natural language processing techniques (e.g., the BERT model) and vectorizes it.

[0515] Step 6:

[0516] The server calculates the similarity between the vectorized search query and the product vectors in the database.

[0517] Input: Search query vector and product feature vector data

[0518] Output: Similarity score

[0519] The server calculates the similarity between the query vector and the product vector using cosine similarity or Euclidean distance.

[0520] Step 7:

[0521] The server lists products in order of similarity and sends the results to the smartphone.

[0522] Input: Similarity score

[0523] Output: List of similar products

[0524] The server lists products with high similarity and sends the results to the terminal.

[0525] Step 8:

[0526] The terminal displays the search results to the user.

[0527] Input: List of similar products

[0528] Output: Displayed product list

[0529] Users can view search results on their smartphones, select products that interest them, and read more details.

[0530] This system allows users to instantly find similar products based on images of products they find in stores.

[0531] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0532] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, the emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0533] System configuration

[0534] 1. Terminal

[0535] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0536] The user interface provides a means for users to enter search queries and also includes an interface for accepting facial recognition and voice input from users.

[0537] 2. Server

[0538] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0539] The search query received from the user interface is received as text, and the user's emotion recognized from the emotion engine is also received.

[0540] The server vectorizes the search query and corrects it based on sentiment data.

[0541] The similarity between the vectorized query and the vectors of past materials is calculated, and materials with high similarity are listed.

[0542] 3. Database

[0543] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[0544] 4. Algorithm

[0545] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for analyzing text queries.

[0546] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0547] Program processing

[0548] 1. Image / slide acquisition and feature extraction

[0549] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0550] The server extracts features from these images and slides. For images, it analyzes and extracts color, shape, texture, and text information from the slides.

[0551] 2. Vectorization

[0552] The server converts the extracted features into vector format, which allows the image data to be represented as a multidimensional vector.

[0553] These vector data are stored in a database.

[0554] 3. Entering a search query and vectorizing it

[0555] A user uses a device to input a text query describing the image or slide atmosphere they want to search for. For example, they might input the text, "I want an image that depicts a bright future."

[0556] The device sends the user's facial expression and voice data to the emotion engine to recognize the user's emotions. For example, if a user searches for "bright future" with a smile, the emotion engine will recognize this as a "positive emotion."

[0557] The server receives the search query and sentiment data, vectorizes the text query, and corrects it based on sentiment.

[0558] 4. Similarity calculation

[0559] The server calculates the similarity between the vectorized query and the vectors in the database, using methods such as cosine similarity or Euclidean distance.

[0560] 5. Displaying the results

[0561] The server lists images and slides in order of similarity and sends the results to the device.

[0562] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0563] Specific examples

[0564] For example, if a user is creating a new project document with the theme of "future cities," they may want to reuse an illustration of a future city from a previous document. The user enters text that describes the atmosphere of a "future city" into the device's search interface and intentionally smiles. The device recognizes the facial expression and sends it to the server as positive emotion data. The server receives the text query and emotion data and generates appropriate search results based on them. This allows the illustration the user is looking for to be efficiently reused from previous documents. This process allows users to find the materials they need quickly and efficiently.

[0565] The processing flow will be explained below.

[0566] Step 1:

[0567] The terminal retrieves images and slides from past documents. Specifically, it accesses the database or file system within the system and reads images and slides contained in planning documents and proposal documents.

[0568] Step 2:

[0569] The server extracts features from the acquired images and slides. The server uses a deep learning model (e.g., Convolutional Neural Network, CNN) to analyze the color, shape, and texture of the images, as well as the text information in the slides, and extracts features.

[0570] Step 3:

[0571] The server vectorizes the extracted features, expresses the features as multidimensional vectors, and generates vectorized data. These vector data are stored in a database.

[0572] Step 4:

[0573] A user uses the device's search interface to enter a text query that specifies the image or slide mood they are looking for, for example, "I want images that depict a bright future."

[0574] Step 5:

[0575] The emotion engine receives user input and analyzes the user's facial expression and voice data to recognize emotions. Specifically, it captures the user's facial expressions through a camera and identifies emotions using a facial expression analysis algorithm. It also captures voice data through a microphone and analyzes emotions using a voice emotion recognition algorithm.

[0576] Step 6:

[0577] The server receives the user's text query and the recognized emotion data from the emotion engine, vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec), and corrects the search query based on the recognized emotion.

[0578] Step 7:

[0579] The server calculates the similarity between the vectorized query and the data vectors in the database, using methods such as cosine similarity or Euclidean distance to measure the distance between each vector and calculate the similarity.

[0580] Step 8:

[0581] The server lists images and slides in order of similarity, compiles the results as search results, generates a search result list, and sends the results to the device.

[0582] Step 9:

[0583] The terminal displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0584] Specific examples

[0585] For example, when creating new planning materials with the theme of "future cities," a user might want to reuse an "illustration of a future city." The user enters text describing the atmosphere of the "future city" into the device's search interface and then captures their own smile with the camera. The emotion engine recognizes the user's smile and generates emotional data representing a "positive emotion." The server receives the text query and the emotional data, and corrects the search query based on both to generate optimal search results. As a result, the "illustration of a future city" desired by the user is efficiently listed and can be reused in creating new planning materials. This process, with the help of the emotion engine, enables searches that more accurately reflect the user's intent.

[0586] Example 2

[0587] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0588] Conventional document search systems not only have difficulty efficiently finding the necessary images and slides from a large number of digital files, but also fail to consider the user's emotions, resulting in search results that often deviate from the user's intentions. Furthermore, feature extraction and search query vectorization rely too heavily on user input, resulting in poor search result accuracy.

[0589] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0590] In this invention, the server includes means for acquiring digital files from past materials, means for extracting features from the acquired digital files and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for recognizing the user's emotions and generating emotion data, means for correcting the vectorized search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past materials and listing materials with high similarity, and means for displaying materials with high similarity. This makes it possible to provide highly accurate search results while taking user emotions into consideration.

[0591] "Past materials" are digital files that were created or collected in the past and registered in the system, and primarily refer to images and slides.

[0592] "Digital files" refers to data stored and exchanged electronically, and generally includes image files and slide files.

[0593] A "feature" is an attribute value or pattern extracted from a digital file, and is information that allows the characteristics of the digital file to be quantified or vectorized.

[0594] "Vectorization" is a technique that represents digital files or text queries as multidimensional numerical vectors, making them easier to process on a computer.

[0595] "Search query" refers to a text input by a user that indicates to the system what they wish to search for.

[0596] "Emotion data" refers to data that expresses the emotional state recognized from the user's facial expressions and voice as numerical values ​​or categories.

[0597] "Similarity" is an index that indicates the closeness or degree of agreement between two vectors (a search query and a digital file's features), and is calculated using methods such as cosine similarity or Euclidean distance.

[0598] "Listing" refers to displaying a list of digital files obtained as search results in order of similarity.

[0599] This invention is a system for efficiently searching for images and slides from past documents, equipped with a function for recognizing user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion recognition engine, and a function for calculating the similarity between vectors. The hardware and software used are described below.

[0600] Hardware and Software Configuration

[0601] 1. Terminal

[0602] Devices used as terminals include PCs, tablets, smartphones, etc.

[0603] The device provides a user interface (UI) for users to enter search queries, which includes not only text input but also facial expression recognition and voice input functions.

[0604] 2. Server

[0605] The server is the main component that processes the data sent from the terminals.

[0606] Deep learning models (e.g., Convolutional Neural Network, CNN) are used to extract features from images and slides.

[0607] Natural language processing techniques (e.g., BERT, Word2Vec) are used to parse and vectorize text queries.

[0608] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0609] 3. Database

[0610] The features of past images and slides are stored in vector format, and the database also includes an index that allows fast and efficient search of vector data.

[0611] Specific operation of the system

[0612] Suppose a user searches for "an illustration that expresses the atmosphere of a futuristic city." The user enters a text query through the device's search interface and intentionally smiles. The device recognizes this facial expression in real time and transmits it to the server as positive emotion data.

[0613] The server vectorizes the text query and the received emotion data, and then modifies the query based on the emotion data. The server then compares the vectors with those of past digital files to identify those with high similarity, using, for example, cosine similarity or Euclidean distance, to generate optimal search results.

[0614] Finally, the server sends the search results to the terminal, which displays them to the user. The user can select images or slides that suit their purpose from the search results and reuse them in new materials. This allows users to quickly and accurately find the materials they need.

[0615] Example prompt sentence:

[0616] Search for "I want an illustration that conveys the atmosphere of a futuristic city."

[0617] Please implement a system that allows users to search for "future cities" with a smile.

[0618] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0619] Step 1: User enters a search query

[0620] The user uses the device to input the desired image or slide content as text into the search interface. For example, they might input "I'm looking for an illustration that evokes the atmosphere of a futuristic city." At this point, the search query text is generated as input data. Once the user has finished entering the query, the device passes this input data to the next processing step.

[0621] Step 2: Recognizing user emotion data by the device

[0622] As the user enters a text query, the device's camera and microphone capture the user's facial expressions and voice. The device then uses an emotion recognition engine to analyze this data and recognize the user's emotion. The emotion data generated is then sent to the next processing step.

[0623] Step 3: Obtaining past documents via terminal

[0624] The device retrieves previously created materials (digital files) from the company's cloud storage or internal server. The input here is past materials retrieved according to conditions specified by the user. The retrieved past materials (images and slides) are sent to the next processing step.

[0625] Step 4: Feature extraction and vectorization on the server

[0626] The server extracts features from images and slides of past documents received from the terminal. A deep learning model (e.g., CNN) is used to extract features. At this time, the input data is past documents that meet the user's requirements, and the output data is the features extracted from these documents. The features are converted into vector format and stored in a database.

[0627] Step 5: Vectorizing search queries by the server

[0628] The server vectorizes the text query entered by the user using natural language processing techniques (e.g., BERT, Word2Vec). Here, the input is the text query and the output is the vectorized query. This vectorized query is sent to the next processing step.

[0629] Step 6: Server-based query refinement based on emotion data

[0630] The server uses the sentiment data to correct the vectorized query. For example, if there is positive sentiment data, it adds a positive weight to the corresponding search query vector. Here, the input is the vectorized query and sentiment data, and the output is the corrected vector query.

[0631] Step 7: Similarity calculation by the server

[0632] The server calculates the similarity between the corrected vector query and the document vectors in the database. The algorithms used include cosine similarity and Euclidean distance. The input data are the corrected query vector and the document vector, and the output is a list of documents with high similarity.

[0633] Step 8: Server lists and sends search results

[0634] As a result of the similarity calculation, the server lists the materials in descending order of similarity and sends it to the terminal. The input here is the result of the similarity calculation, and the output is a list of search results. The list of listed materials is sent to the terminal.

[0635] Step 9: Device displays search results and user selection

[0636] The terminal displays the received search results on the user interface. The user can select images or slides that suit their purpose from this list. The input here is the list of search results, and the output is the materials selected by the user. The user can reuse the selected materials in new project materials to create.

[0637] (Application example 2)

[0638] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0639] When searching for images or slides from past documents, it is difficult to obtain appropriate results that reflect the user's emotions. In particular, if search results are generated only by simple text matching or vector comparison without considering the user's emotions when searching, the optimal results may not be provided according to the user's expectations.

[0640] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0641] In this invention, the server includes means for extracting and vectorizing acquired images and slides as features, means for vectorizing search queries entered by users, means for recognizing user emotions and correcting the search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past documents and listing documents with high similarity, and means for receiving user emotions as feedback and reflecting them in the next search, thereby making it possible to provide optimal search results that reflect the user's emotions.

[0642] - "Historical materials" refers to data such as digital documents, images, and slides that have been collected and stored to date.

[0643] "Image" refers to any visual data displayed or stored by a digital device.

[0644] "Slides" refers to individual visual pages used in presentations and reports.

[0645] "Means of acquisition" refers to a method or device for capturing data of historical materials from a designated location.

[0646] "Means for extracting features" refers to a method or device that analyzes the visual and content elements of an image or slide and converts them into numerical data.

[0647] "Vectorization means" refers to a method or device that formalizes extracted features as multidimensional vectors.

[0648] "Search query" refers to a string of characters or spoken instructions entered by a user to indicate the information or material they are seeking.

[0649] "Means for recognizing emotions" refers to technology or devices that determine a user's psychological state at that time from facial expressions and voice data.

[0650] "Means for correcting based on emotional data" refers to technologies or devices that adjust or transform search queries taking into account the emotional state of the user.

[0651] The term "means for calculating similarity" refers to an algorithm or device for numerically evaluating the similarity between vector-format data.

[0652] The "listing means" refers to a method or device for preferentially extracting data with high similarity and displaying it in a list format.

[0653] "Displaying means" refers to a method or device for visually presenting search results to a user.

[0654] "Means for receiving feedback" refers to a method or device for incorporating user evaluations and opinions into the system.

[0655] A "deep learning model" refers to an algorithm that uses an advanced neural network to extract and learn features.

[0656] "Natural language processing technology" refers to all technologies for analyzing and understanding natural language texts that humans use on a daily basis.

[0657] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. The system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0658] System configuration

[0659] 1. Terminal

[0660] The terminal retrieves previously created planning and proposal materials and sends them to the system. The user interface provides a means for users to input search queries. It also includes an interface that recognizes the user's facial expressions and receives voice input. Specifically, a smartphone or smart glasses can function as the terminal.

[0661] 2. Server

[0662] The server extracts features from images and slides obtained from the device and vectorizes them. A deep learning model (e.g., CNN) is used to extract features, and natural language processing technology (e.g., BERT, Word2Vec) is used to analyze text queries. The server receives the search query as text from the user interface, as well as the user's emotions recognized by the emotion engine. The server vectorizes the search query and corrects it based on the emotion data. The server calculates the similarity between the vectorized query and the vectors of past materials and lists materials with high similarity.

[0663] 3. Database

[0664] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[0665] 4. Algorithm

[0666] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for text query analysis. The emotion engine uses technologies to recognize user facial expressions and identify emotions from voice (e.g., facial expression analysis, voice emotion recognition).

[0667] Program processing

[0668] The server uses a deep learning model to extract features from images and slides sent from the device and vectorizes them. The device receives the search query entered by the user through the user interface as text and recognizes the user's emotions using an emotion engine. The server receives this text query and emotion data, vectorizes the query using natural language processing technology, and corrects it based on the emotion data. The server then calculates the similarity between the vectorized query and the vectors of past documents and lists documents with high similarity. Finally, the search results are displayed to the user via the device. The user can select images or slides that match their purpose from the displayed results and reuse them in creating documents.

[0669] Specific examples

[0670] For example, imagine a user is searching for "new sneakers" in a physical store. The user launches a smartphone app, enters "comfortable sneakers" in the search field, and intentionally smiles. The app recognizes this facial expression and sends it to the server as a positive emotion.

[0671] Example prompt (input to generative AI model):

[0672] A user is searching for "comfortable sneakers." The emotion engine recognizes the user's smile and confirms that the emotion is positive. Find the most suitable sneakers based on past purchase history and product reviews. Also, display related product recommendations in the search results.

[0673] This process allows users to quickly find the product that best suits their emotions, improving the shopping experience in physical stores. Specifically, it lists comfortable sneakers and related products that users have previously purchased, and allows users to try them on and provide feedback to further improve the service.

[0674] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0675] Step 1:

[0676] The terminal retrieves images and slides from past documents and sends this data to the system. Specifically, it retrieves previously created planning and proposal documents from a database. The input is past document data, and the output is data sent to the server.

[0677] Step 2:

[0678] The server uses a deep learning model (e.g., CNN) to extract features from acquired images and slides and vectorize them. Specifically, it analyzes and extracts image color, shape, texture, and text information from slides. The input is image or slide data, and the output is feature data in vector format.

[0679] Step 3:

[0680] The user uses the device to input the mood of the image or slide they want to search for as a text query. At this time, the device also sends the user's facial expressions and voice input to the emotion engine to recognize emotions. The input is the user's text query and emotion data, and the output is data sent from the device to the server.

[0681] Step 4:

[0682] The server analyzes and vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec). It then corrects the vector based on emotion data from the emotion engine. The input is the text query and emotion data, and the output is the corrected vector-formatted search query.

[0683] Step 5:

[0684] The server calculates the similarity between the vectorized search query and the vectors of past documents using methods such as cosine similarity or Euclidean distance. The input is the corrected search query vector and the vectors of past documents, and the output is a list of documents with high similarity.

[0685] Step 6:

[0686] The server lists the search results in order of similarity and sends them to the terminal. The input is a list of similarities between documents, and the output is the transmission of the search results to the terminal.

[0687] Step 7:

[0688] The terminal displays the search results to the user. The user selects images and slides that fit the purpose from the displayed list and reuses them in creating materials. The input is the search results sent from the server, and the output is a visual presentation to the user.

[0689] Step 8:

[0690] The user inputs feedback such as satisfaction level for the selected result, and the terminal transmits this feedback to the server. The input is the user's feedback data, and the output is data transmission to the server.

[0691] Step 9:

[0692] The server reflects the received feedback in the emotion engine to improve the accuracy of the next search results. The input is the feedback data, and the output is an improved emotion model.

[0693] 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 a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0694] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0695] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[0696] [Third embodiment]

[0697] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0698] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[0699] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0700] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[0701] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0702] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0703] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0704] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0705] The specific processing program 56 is an example of a "program" according to the technology of the present 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.

[0706] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0707] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. 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 process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0708] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. 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."

[0709] This invention relates to a system for efficiently searching for images and slides from past documents. This system includes a database for managing past documents, a user interface for inputting search queries, an algorithm for extracting and vectorizing features, and a function for calculating the similarity between the vectors of the documents and the search query.

[0710] System configuration

[0711] 1. Terminal

[0712] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0713] The terminal is provided with an input interface for the user to input a search query.

[0714] 2. Server

[0715] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0716] The server receives a search query from a user and vectorizes the query.

[0717] The server calculates the similarity between the vectorized query and the vectors of images or slides stored in the database.

[0718] Based on the results of the similarity calculation, the materials are listed in descending order of similarity.

[0719] 3. Database

[0720] The database stores vectorized data of past images and slides.

[0721] 4. Algorithm

[0722] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used to analyze and vectorize text queries.

[0723] Program processing

[0724] 1. Image / slide acquisition and feature extraction

[0725] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0726] The server extracts features from these images and slides, analyzing color, shape, texture, and text in the case of diagram images.

[0727] 2. Vectorization

[0728] The server converts the extracted features into vector format, for example, by using a deep learning model to convert image data into a multidimensional vector.

[0729] The extracted vector data is stored in a database.

[0730] 3. Entering a search query and vectorizing it

[0731] The user uses the input interface of the device to enter specific text describing the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0732] The server receives this text query and vectorizes it using natural language processing techniques, such as the BERT model, to convert the text into multidimensional vectors.

[0733] 4. Similarity calculation

[0734] The server calculates the similarity between the vectorized query and the vectors in the database. It measures the distance using cosine similarity or Euclidean distance, and calculates the similarity.

[0735] 5. Displaying the results

[0736] The server lists images and slides in order of similarity and sends the results to the device.

[0737] The terminal displays the results to the user, who can then check and select the images and slides they need.

[0738] Specific examples

[0739] For example, if a user is creating new planning materials on the theme of "future cities," they may want to reuse an "illustration of a future city" that was included in a previous document. The user enters text that describes the atmosphere of the "future city" into the search interface on their device. The server converts this text into vectors and calculates the similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can then select the "illustration of a future city" from the list and reuse it in the new document. This process allows users to find the materials they need quickly and efficiently.

[0740] The processing flow will be explained below.

[0741] Step 1:

[0742] The device retrieves images and slides from past documents, and loads multiple files containing graphs showing a company's growth prospects and illustrations of future cities.

[0743] Step 2:

[0744] The server extracts the features of the acquired images and slides. Using a deep learning model (e.g., Convolutional Neural Network, CNN), the server analyzes and extracts important features such as the color, shape, and texture of the images, as well as text information within the slides.

[0745] Step 3:

[0746] The server vectorizes the extracted features. The features are expressed as multidimensional vectors, and vectorized data is generated. This vector data is stored in a database.

[0747] Step 4:

[0748] A user uses a search interface on a device to enter a text query that specifies the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[0749] Step 5:

[0750] The server analyzes and vectorizes the text query received from the user using natural language processing techniques. Specifically, it uses text encoding techniques such as BERT and Word2Vec to convert the text query into a multidimensional vector.

[0751] Step 6:

[0752] The server calculates the similarity between the vectorized query and the vector data of past documents in the database. It measures the distance between each vector using methods such as cosine similarity and Euclidean distance to calculate the similarity.

[0753] Step 7:

[0754] The server lists images and slides in descending order of similarity, generating a list of documents with the highest similarity first, and compiling this list as search results.

[0755] Step 8:

[0756] The server generates a search result list and sends it to the device, containing the images and slides that best match the user's search query.

[0757] Step 9:

[0758] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0759] In this way, even if a user cannot remember specific keywords, they can efficiently search for and reuse the necessary materials based on the atmosphere of the images and slides.

[0760] Example 1

[0761] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0762] Efficiently searching and retrieving necessary images and slides from past documents is a time-consuming and labor-intensive process. It is especially difficult to quickly find the desired content when a large amount of documents are stored in a database. Furthermore, accurate interpretation of user search queries and provision of optimal documents requires advanced technology.

[0763] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0764] In this invention, the server includes a means for uploading past materials specified by a user, a means for using a deep learning model to extract features of the uploaded materials, and a means for analyzing and vectorizing search queries using a natural language processing model, thereby enabling efficient search for necessary images and slides from past materials and providing the necessary materials with high accuracy.

[0765] "Images and slides" are visual content stored as past documents, and refer to graphics and presentation materials that provide concrete visual information.

[0766] "Features" are data that numerically represent specific properties or patterns extracted from digital data such as images or slides.

[0767] "Vectorization" is the process of representing features as vectors in a multidimensional space, which makes it easier to calculate similarities between data.

[0768] A "search query" is a text input used by a user to search for specific information, thereby providing guidance to the system in locating appropriate materials.

[0769] "Natural language processing technology" refers to a computational technology for analyzing and understanding human language, and in particular to the technology for semantically processing input text.

[0770] A "deep learning model" is a type of machine learning model that uses a multi-layer neural network to automatically learn complex patterns and features from data.

[0771] "Cosine similarity" is a method for calculating the similarity between two vectors, and uses the cosine value of the angle between the vectors.

[0772] "Euclidean distance" is a method for calculating the straight-line distance between two points (vectors). The smaller the distance, the more similar the two points (vectors) are considered to be.

[0773] A "database" is a system for efficiently storing, managing, and searching structured data.

[0774] This invention is a system that efficiently searches for images and slides from past documents and provides them to users. This system is composed of terminals, servers, databases, deep learning models, and natural language processing models.

[0775] First, the device retrieves images and slides from past documents, including hardware such as online storage and local file systems. The user can then specify the documents using the device's input interface and import them into the system.

[0776] The captured images and slides are then processed on a server where features are extracted. The server uses deep learning libraries (e.g., TensorFlow and PyTorch) to analyze image features such as color, shape, texture, and text. These features are represented as multidimensional vectors.

[0777] The server then uses a pre-trained deep learning model (e.g., InceptionV3 or ResNet) to convert these extracted features into vector format, which is then stored in a database (e.g., MongoDB or MySQL).

[0778] When a user enters a search query, they use the input interface of their device. For example, they enter specific text such as "illustration of a futuristic city" and click the search button. This text query is then vectorized by the server using a natural language processing model (e.g., BERT or Word2Vec).

[0779] The vectorized search query is used to calculate the similarity between the vector data of images and slides in the database. The server uses cosine similarity or Euclidean distance for this similarity calculation. Based on the calculated similarity, materials with high similarity are listed.

[0780] Finally, the server sends a list of materials with high similarity to the terminal, which displays it to the user. The user can then select and reuse the images and slides they need from the displayed materials. This system allows users to quickly and efficiently find the materials they need.

[0781] For example, consider a user creating new project materials on the theme of "future cities." The user enters "illustration of a future city" into the search interface on their device and clicks the search button. The server vectorizes this text and calculates its similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can select the "illustration of a future city" from the list and reuse it in the new material. This process allows users to quickly and efficiently find the materials they need.

[0782] Prompt Sentence Examples

[0783] "I want to create a planning document for a future city. Please search past documents for images and slides on the theme of future cities."

[0784] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0785] Step 1: Acquire images and slides

[0786] The device retrieves images and slides from past documents. The user uses a device such as a PC or smartphone to select a specific file from online storage or the local file system and clicks the upload button. In this step, the user can upload a "graph showing a company's growth prospects" or "illustration of a future city," and these files are sent to the system. The input data are the image or slide files, and the output data is a notification that the upload has been completed to the server.

[0787] Step 2: Feature extraction

[0788] The server analyzes images and slides received from the device and extracts features. Deep learning libraries (e.g., TensorFlow and PyTorch) are used for extraction. The image's color distribution, shape, texture, and included text are automatically analyzed. The input data are the uploaded image or slide files, and the output data is a dataset containing the extracted features.

[0789] Step 3: Vectorization

[0790] The server converts the features into multidimensional vectors. A pre-trained deep learning model (e.g., InceptionV3 or ResNet) is used to map the extracted features into a high-dimensional vector space. The input data are the features extracted in step 2, and the output data is data represented as multidimensional vectors.

[0791] Step 4: Saving to the Database

[0792] The server stores the vectorized data in a database (e.g., MongoDB or MySQL). At this time, metadata (e.g., title, date, category, etc.) of the corresponding image or slide is also stored for each vector. The input data is data and metadata expressed as multidimensional vectors, and the output data is vector data stored in the database.

[0793] Step 5: Enter a search query

[0794] The user enters a search query using the search interface on the device. For example, they enter specific keywords such as "illustration of a futuristic city" or "image showing a bright future," and clicks the search button. The input data is the text of the search query, and the output data is a notification that the query has been sent to the server.

[0795] Step 6: Vectorizing the query

[0796] The server vectorizes the search query received from the user using a natural language processing model (e.g., BERT or Word2Vec). It captures the meaning of the query and represents the text as a multidimensional vector. The input data is the search query text, and the output data is the vectorized query.

[0797] Step 7: Similarity calculation

[0798] The server calculates the similarity between the vectorized query and the vector data in the database using algorithms such as cosine similarity or Euclidean distance. The input data are the vectorized query and the vector data in the database, and the output data is the similarity score.

[0799] Step 8: Viewing the results

[0800] The server lists images and slides in descending order of similarity based on the similarity score and sends the list to the terminal. The terminal displays the results to the user, who can then select the images or slides they need from the displayed list. The input data is the similarity list generated by the server, and the output data is the results displayed on the user's terminal.

[0801] (Application example 1)

[0802] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0803] In today's brick-and-mortar stores, it is difficult for consumers to efficiently search for products based on their preferences and past purchase history. It is also difficult for them to instantly understand the vast amount of product information available in stores and make appropriate selections. Therefore, there is a need for a method that allows consumers to find products more quickly and accurately.

[0804] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0805] In this invention, the server includes means for acquiring images and slides from past materials, means for extracting features from the acquired images and slides and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for calculating the similarity between the vectorized search query and the vectors of the past materials and listing materials with high similarity, means for displaying materials with high similarity, means for searching for retail products based on images, means for vectorizing an image taken as the search query and calculating the similarity with product vectors in a database, and means for listing and displaying products with high similarity. This allows a user to instantly find similar products based on images of products they found in a store.

[0806] "Past materials" refers to previously created images, slides, and document data.

[0807] "Features of images and slides" refer to important elements that make up an image, such as color, shape, texture, and text, extracted from the image or slide.

[0808] "Vectorization" is the process of converting features into points in a multidimensional space.

[0809] A "search query" is the text or image information that a user enters into the system.

[0810] "Similarity" is an index that measures the similarity between two vectors, and there are calculation methods such as cosine similarity and Euclidean distance.

[0811] A "database" is a digital warehouse that stores and manages vectorized data extracted from past documents.

[0812] "Natural language processing technology" is a technology that understands and analyzes text data and converts it into a format that can be processed by machines.

[0813] A "deep learning model" is an artificial intelligence technology that automatically learns features using large amounts of data and makes highly accurate predictions and classifications.

[0814] "Retail" is a general term that refers to retail trade and brick-and-mortar sales.

[0815] A "means for calculating similarity" refers to a method or technique for comparing a vectorized query with a vector in a database and measuring the similarity between them.

[0816] This invention relates to a system for improving the shopping experience in brick-and-mortar stores. The system aims to efficiently search for and present related and recommended products to users based on their past purchase history and images and descriptions of products seen in stores.

[0817] The system uses a smartphone as hardware and databases (e.g., MongoDB), deep learning models (e.g., CNN using TensorFlow), and natural language processing models (e.g., BERT) as software. It uses Node.js and Express on the server side and React Native on the front end.

[0818] A user uses the camera function of their smartphone to take a picture of a product they see in a store. This image is sent to the server via the device. The server extracts features from the received image and vectorizes them using a deep learning model. These features include the product's shape, color, texture, etc.

[0819] The server stores the vectorized feature data in a database. Furthermore, the server receives search queries (e.g., "red dress" or "sports shoes") entered by users as text and vectorizes them using natural language processing techniques. The BERT model is used for this vectorization.

[0820] The server calculates the similarity between the vectorized search query and the product vectors in the database. This similarity calculation uses cosine similarity or Euclidean distance. The server then lists the products in descending order of similarity and sends the results to the smartphone.

[0821] The smartphone displays products sorted by similarity, and users can select the product they are interested in and check its details. This allows users to instantly find similar products based on the image of the product they found in the store.

[0822] For example, if a user is looking for a "red dress" in a store, they can take a photo of the product with their smartphone and enter the keyword "red dress." The server analyzes the features based on the image and keyword, and vectorizes them using a deep learning model and natural language processing technology. The server then calculates the similarity with products in the database and displays the results on the user's smartphone.

[0823] Examples of prompts include:

[0824] I'm looking for a "red dress." Show me products similar to this in-store image.

[0825] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0826] Step 1:

[0827] A user takes a photo of a product in a store using the camera on their smartphone.

[0828] Input: Image taken inside the store

[0829] Output: Captured image data

[0830] Image data is sent to the server, which receives the data and begins processing it.

[0831] Step 2:

[0832] The server extracts features from the received image data.

[0833] Input: Photographed image data

[0834] Output: Feature vector

[0835] The server uses a deep learning model (e.g., CNN using TensorFlow) to extract features such as color, shape, and texture from the image.

[0836] Step 3:

[0837] The server vectorizes the extracted features.

[0838] Input: extracted features

[0839] Output: A vector of features

[0840] The server converts these features into multidimensional vectors, which are then stored in a MongoDB database.

[0841] Step 4:

[0842] A user inputs a search query through an input interface of a smartphone.

[0843] Input: Search query (e.g. "red dress")

[0844] Output: Query text

[0845] The terminal transmits this text data to the server.

[0846] Step 5:

[0847] The text data of the search query received by the server is vectorized.

[0848] Input: Text data of the search query

[0849] Output: A vector of text

[0850] The server analyzes this text using natural language processing techniques (e.g., the BERT model) and vectorizes it.

[0851] Step 6:

[0852] The server calculates the similarity between the vectorized search query and the product vectors in the database.

[0853] Input: Search query vector and product feature vector data

[0854] Output: Similarity score

[0855] The server calculates the similarity between the query vector and the product vector using cosine similarity or Euclidean distance.

[0856] Step 7:

[0857] The server lists products in order of similarity and sends the results to the smartphone.

[0858] Input: Similarity score

[0859] Output: List of similar products

[0860] The server lists products with high similarity and sends the results to the terminal.

[0861] Step 8:

[0862] The terminal displays the search results to the user.

[0863] Input: List of similar products

[0864] Output: Displayed product list

[0865] Users can view search results on their smartphones, select products that interest them, and read more details.

[0866] This system allows users to instantly find similar products based on images of products they find in stores.

[0867] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0868] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, the emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0869] System configuration

[0870] 1. Terminal

[0871] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[0872] The user interface provides a means for users to enter search queries and also includes an interface for accepting facial recognition and voice input from users.

[0873] 2. Server

[0874] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[0875] The search query received from the user interface is received as text, and the user's emotion recognized from the emotion engine is also received.

[0876] The server vectorizes the search query and corrects it based on sentiment data.

[0877] The similarity between the vectorized query and the vectors of past materials is calculated, and materials with high similarity are listed.

[0878] 3. Database

[0879] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[0880] 4. Algorithm

[0881] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for analyzing text queries.

[0882] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0883] Program processing

[0884] 1. Image / slide acquisition and feature extraction

[0885] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[0886] The server extracts features from these images and slides. For images, it analyzes and extracts color, shape, texture, and text information from the slides.

[0887] 2. Vectorization

[0888] The server converts the extracted features into vector format, which allows the image data to be represented as a multidimensional vector.

[0889] These vector data are stored in a database.

[0890] 3. Entering a search query and vectorizing it

[0891] A user uses a device to input a text query describing the image or slide atmosphere they want to search for. For example, they might input the text, "I want an image that depicts a bright future."

[0892] The device sends the user's facial expression and voice data to the emotion engine to recognize the user's emotions. For example, if a user searches for "bright future" with a smile, the emotion engine will recognize this as a "positive emotion."

[0893] The server receives the search query and sentiment data, vectorizes the text query, and corrects it based on sentiment.

[0894] 4. Similarity calculation

[0895] The server calculates the similarity between the vectorized query and the vectors in the database, using methods such as cosine similarity or Euclidean distance.

[0896] 5. Displaying the results

[0897] The server lists images and slides in order of similarity and sends the results to the device.

[0898] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0899] Specific examples

[0900] For example, if a user is creating a new project document with the theme of "future cities," they may want to reuse an illustration of a future city from a previous document. The user enters text that describes the atmosphere of a "future city" into the device's search interface and intentionally smiles. The device recognizes the facial expression and sends it to the server as positive emotion data. The server receives the text query and emotion data and generates appropriate search results based on them. This allows the illustration the user is looking for to be efficiently reused from previous documents. This process allows users to find the materials they need quickly and efficiently.

[0901] The processing flow will be explained below.

[0902] Step 1:

[0903] The terminal retrieves images and slides from past documents. Specifically, it accesses the database or file system within the system and reads images and slides contained in planning documents and proposal documents.

[0904] Step 2:

[0905] The server extracts features from the acquired images and slides. The server uses a deep learning model (e.g., Convolutional Neural Network, CNN) to analyze the color, shape, and texture of the images, as well as the text information in the slides, and extracts features.

[0906] Step 3:

[0907] The server vectorizes the extracted features, expresses the features as multidimensional vectors, and generates vectorized data. These vector data are stored in a database.

[0908] Step 4:

[0909] A user uses the device's search interface to enter a text query that specifies the image or slide mood they are looking for, for example, "I want images that depict a bright future."

[0910] Step 5:

[0911] The emotion engine receives user input and analyzes the user's facial expression and voice data to recognize emotions. Specifically, it captures the user's facial expressions through a camera and identifies emotions using a facial expression analysis algorithm. It also captures voice data through a microphone and analyzes emotions using a voice emotion recognition algorithm.

[0912] Step 6:

[0913] The server receives the user's text query and the recognized emotion data from the emotion engine, vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec), and corrects the search query based on the recognized emotion.

[0914] Step 7:

[0915] The server calculates the similarity between the vectorized query and the data vectors in the database, using methods such as cosine similarity or Euclidean distance to measure the distance between each vector and calculate the similarity.

[0916] Step 8:

[0917] The server lists images and slides in order of similarity, compiles the results as search results, generates a search result list, and sends the results to the device.

[0918] Step 9:

[0919] The terminal displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[0920] Specific examples

[0921] For example, when creating new planning materials with the theme of "future cities," a user might want to reuse an "illustration of a future city." The user enters text describing the atmosphere of the "future city" into the device's search interface and then captures their own smile with the camera. The emotion engine recognizes the user's smile and generates emotional data representing a "positive emotion." The server receives the text query and the emotional data, and corrects the search query based on both to generate optimal search results. As a result, the "illustration of a future city" desired by the user is efficiently listed and can be reused in creating new planning materials. This process, with the help of the emotion engine, enables searches that more accurately reflect the user's intent.

[0922] Example 2

[0923] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0924] Conventional document search systems not only have difficulty efficiently finding the necessary images and slides from a large number of digital files, but also fail to consider the user's emotions, resulting in search results that often deviate from the user's intentions. Furthermore, feature extraction and search query vectorization rely too heavily on user input, resulting in poor search result accuracy.

[0925] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0926] In this invention, the server includes means for acquiring digital files from past materials, means for extracting features from the acquired digital files and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for recognizing the user's emotions and generating emotion data, means for correcting the vectorized search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past materials and listing materials with high similarity, and means for displaying materials with high similarity. This makes it possible to provide highly accurate search results while taking user emotions into consideration.

[0927] "Past materials" are digital files that were created or collected in the past and registered in the system, and primarily refer to images and slides.

[0928] "Digital files" refers to data stored and exchanged electronically, and generally includes image files and slide files.

[0929] A "feature" is an attribute value or pattern extracted from a digital file, and is information that allows the characteristics of the digital file to be quantified or vectorized.

[0930] "Vectorization" is a technique that represents digital files or text queries as multidimensional numerical vectors, making them easier to process on a computer.

[0931] "Search query" refers to a text input by a user that indicates to the system what they wish to search for.

[0932] "Emotion data" refers to data that expresses the emotional state recognized from the user's facial expressions and voice as numerical values ​​or categories.

[0933] "Similarity" is an index that indicates the closeness or degree of agreement between two vectors (a search query and a digital file's features), and is calculated using methods such as cosine similarity or Euclidean distance.

[0934] "Listing" refers to displaying a list of digital files obtained as search results in order of similarity.

[0935] This invention is a system for efficiently searching for images and slides from past documents, equipped with a function for recognizing user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion recognition engine, and a function for calculating the similarity between vectors. The hardware and software used are described below.

[0936] Hardware and Software Configuration

[0937] 1. Terminal

[0938] Devices used as terminals include PCs, tablets, smartphones, etc.

[0939] The device provides a user interface (UI) for users to enter search queries, which includes not only text input but also facial expression recognition and voice input functions.

[0940] 2. Server

[0941] The server is the main component that processes the data sent from the terminals.

[0942] Deep learning models (e.g., Convolutional Neural Network, CNN) are used to extract features from images and slides.

[0943] Natural language processing techniques (e.g., BERT, Word2Vec) are used to parse and vectorize text queries.

[0944] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[0945] 3. Database

[0946] The features of past images and slides are stored in vector format, and the database also includes an index that allows fast and efficient search of vector data.

[0947] Specific operation of the system

[0948] Suppose a user searches for "an illustration that expresses the atmosphere of a futuristic city." The user enters a text query through the device's search interface and intentionally smiles. The device recognizes this facial expression in real time and transmits it to the server as positive emotion data.

[0949] The server vectorizes the text query and the received emotion data, and then modifies the query based on the emotion data. The server then compares the vectors with those of past digital files to identify those with high similarity, using, for example, cosine similarity or Euclidean distance, to generate optimal search results.

[0950] Finally, the server sends the search results to the terminal, which displays them to the user. The user can select images or slides that suit their purpose from the search results and reuse them in new materials. This allows users to quickly and accurately find the materials they need.

[0951] Example prompt sentence:

[0952] Search for "I want an illustration that conveys the atmosphere of a futuristic city."

[0953] Please implement a system that allows users to search for "future cities" with a smile.

[0954] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0955] Step 1: User enters a search query

[0956] The user uses the device to input the desired image or slide content as text into the search interface. For example, they might input "I'm looking for an illustration that evokes the atmosphere of a futuristic city." At this point, the search query text is generated as input data. Once the user has finished entering the query, the device passes this input data to the next processing step.

[0957] Step 2: Recognizing user emotion data by the device

[0958] As the user enters a text query, the device's camera and microphone capture the user's facial expressions and voice. The device then uses an emotion recognition engine to analyze this data and recognize the user's emotion. The emotion data generated is then sent to the next processing step.

[0959] Step 3: Obtaining past documents via terminal

[0960] The device retrieves previously created materials (digital files) from the company's cloud storage or internal server. The input here is past materials retrieved according to conditions specified by the user. The retrieved past materials (images and slides) are sent to the next processing step.

[0961] Step 4: Feature extraction and vectorization on the server

[0962] The server extracts features from images and slides of past documents received from the terminal. A deep learning model (e.g., CNN) is used to extract features. At this time, the input data is past documents that meet the user's requirements, and the output data is the features extracted from these documents. The features are converted into vector format and stored in a database.

[0963] Step 5: Vectorizing search queries by the server

[0964] The server vectorizes the text query entered by the user using natural language processing techniques (e.g., BERT, Word2Vec). Here, the input is the text query and the output is the vectorized query. This vectorized query is sent to the next processing step.

[0965] Step 6: Server-based query refinement based on emotion data

[0966] The server uses the sentiment data to correct the vectorized query. For example, if there is positive sentiment data, it adds a positive weight to the corresponding search query vector. Here, the input is the vectorized query and sentiment data, and the output is the corrected vector query.

[0967] Step 7: Similarity calculation by the server

[0968] The server calculates the similarity between the corrected vector query and the document vectors in the database. The algorithms used include cosine similarity and Euclidean distance. The input data are the corrected query vector and the document vector, and the output is a list of documents with high similarity.

[0969] Step 8: Server lists and sends search results

[0970] As a result of the similarity calculation, the server lists the materials in descending order of similarity and sends it to the terminal. The input here is the result of the similarity calculation, and the output is a list of search results. The list of listed materials is sent to the terminal.

[0971] Step 9: Device displays search results and user selection

[0972] The terminal displays the received search results on the user interface. The user can select images or slides that suit their purpose from this list. The input here is the list of search results, and the output is the materials selected by the user. The user can reuse the selected materials in new project materials to create.

[0973] (Application example 2)

[0974] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0975] When searching for images or slides from past documents, it is difficult to obtain appropriate results that reflect the user's emotions. In particular, if search results are generated only by simple text matching or vector comparison without considering the user's emotions when searching, the optimal results may not be provided according to the user's expectations.

[0976] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0977] In this invention, the server includes means for extracting and vectorizing acquired images and slides as features, means for vectorizing search queries entered by users, means for recognizing user emotions and correcting the search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past documents and listing documents with high similarity, and means for receiving user emotions as feedback and reflecting them in the next search, thereby making it possible to provide optimal search results that reflect the user's emotions.

[0978] - "Historical materials" refers to data such as digital documents, images, and slides that have been collected and stored to date.

[0979] "Image" refers to any visual data displayed or stored by a digital device.

[0980] "Slides" refers to individual visual pages used in presentations and reports.

[0981] "Means of acquisition" refers to a method or device for capturing data of historical materials from a designated location.

[0982] "Means for extracting features" refers to a method or device that analyzes the visual and content elements of an image or slide and converts them into numerical data.

[0983] "Vectorization means" refers to a method or device that formalizes extracted features as multidimensional vectors.

[0984] "Search query" refers to a string of characters or spoken instructions entered by a user to indicate the information or material they are seeking.

[0985] "Means for recognizing emotions" refers to technology or devices that determine a user's psychological state at that time from facial expressions and voice data.

[0986] "Means for correcting based on emotional data" refers to technologies or devices that adjust or transform search queries taking into account the emotional state of the user.

[0987] The term "means for calculating similarity" refers to an algorithm or device for numerically evaluating the similarity between vector-format data.

[0988] The "listing means" refers to a method or device for preferentially extracting data with high similarity and displaying it in a list format.

[0989] "Displaying means" refers to a method or device for visually presenting search results to a user.

[0990] "Means for receiving feedback" refers to a method or device for incorporating user evaluations and opinions into the system.

[0991] A "deep learning model" refers to an algorithm that uses an advanced neural network to extract and learn features.

[0992] "Natural language processing technology" refers to all technologies for analyzing and understanding natural language texts that humans use on a daily basis.

[0993] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. The system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[0994] System configuration

[0995] 1. Terminal

[0996] The terminal retrieves previously created planning and proposal materials and sends them to the system. The user interface provides a means for users to input search queries. It also includes an interface that recognizes the user's facial expressions and receives voice input. Specifically, a smartphone or smart glasses can function as the terminal.

[0997] 2. Server

[0998] The server extracts features from images and slides obtained from the device and vectorizes them. A deep learning model (e.g., CNN) is used to extract features, and natural language processing technology (e.g., BERT, Word2Vec) is used to analyze text queries. The server receives the search query as text from the user interface, as well as the user's emotions recognized by the emotion engine. The server vectorizes the search query and corrects it based on the emotion data. The server calculates the similarity between the vectorized query and the vectors of past materials and lists materials with high similarity.

[0999] 3. Database

[1000] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[1001] 4. Algorithm

[1002] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for text query analysis. The emotion engine uses technologies to recognize user facial expressions and identify emotions from voice (e.g., facial expression analysis, voice emotion recognition).

[1003] Program processing

[1004] The server uses a deep learning model to extract features from images and slides sent from the device and vectorizes them. The device receives the search query entered by the user through the user interface as text and recognizes the user's emotions using an emotion engine. The server receives this text query and emotion data, vectorizes the query using natural language processing technology, and corrects it based on the emotion data. The server then calculates the similarity between the vectorized query and the vectors of past documents and lists documents with high similarity. Finally, the search results are displayed to the user via the device. The user can select images or slides that match their purpose from the displayed results and reuse them in creating documents.

[1005] Specific examples

[1006] For example, imagine a user is searching for "new sneakers" in a physical store. The user launches a smartphone app, enters "comfortable sneakers" in the search field, and intentionally smiles. The app recognizes this facial expression and sends it to the server as a positive emotion.

[1007] Example prompt (input to generative AI model):

[1008] A user is searching for "comfortable sneakers." The emotion engine recognizes the user's smile and confirms that the emotion is positive. Find the most suitable sneakers based on past purchase history and product reviews. Also, display related product recommendations in the search results.

[1009] This process allows users to quickly find the product that best suits their emotions, improving the shopping experience in physical stores. Specifically, it lists comfortable sneakers and related products that users have previously purchased, and allows users to try them on and provide feedback to further improve the service.

[1010] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1011] Step 1:

[1012] The terminal retrieves images and slides from past documents and sends this data to the system. Specifically, it retrieves previously created planning and proposal documents from a database. The input is past document data, and the output is data sent to the server.

[1013] Step 2:

[1014] The server uses a deep learning model (e.g., CNN) to extract features from acquired images and slides and vectorize them. Specifically, it analyzes and extracts image color, shape, texture, and text information from slides. The input is image or slide data, and the output is feature data in vector format.

[1015] Step 3:

[1016] The user uses the device to input the mood of the image or slide they want to search for as a text query. At this time, the device also sends the user's facial expressions and voice input to the emotion engine to recognize emotions. The input is the user's text query and emotion data, and the output is data sent from the device to the server.

[1017] Step 4:

[1018] The server analyzes and vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec). It then corrects the vector based on emotion data from the emotion engine. The input is the text query and emotion data, and the output is the corrected vector-formatted search query.

[1019] Step 5:

[1020] The server calculates the similarity between the vectorized search query and the vectors of past documents using methods such as cosine similarity or Euclidean distance. The input is the corrected search query vector and the vectors of past documents, and the output is a list of documents with high similarity.

[1021] Step 6:

[1022] The server lists the search results in order of similarity and sends them to the terminal. The input is a list of similarities between documents, and the output is the transmission of the search results to the terminal.

[1023] Step 7:

[1024] The terminal displays the search results to the user. The user selects images and slides that fit the purpose from the displayed list and reuses them in creating materials. The input is the search results sent from the server, and the output is a visual presentation to the user.

[1025] Step 8:

[1026] The user inputs feedback such as satisfaction level for the selected result, and the terminal transmits this feedback to the server. The input is the user's feedback data, and the output is data transmission to the server.

[1027] Step 9:

[1028] The server reflects the received feedback in the emotion engine to improve the accuracy of the next search results. The input is the feedback data, and the output is an improved emotion model.

[1029] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1030] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1031] 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 the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1032] [Fourth embodiment]

[1033] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1034] 7, a 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.

[1035] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1036] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1037] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1038] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1039] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1040] The control object 443 includes a display device, LEDs in the eyes, and motors for driving 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 emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1041] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1042] The specific processing program 56 is an example of a "program" according to the technology of the present 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.

[1043] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1044] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. 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 process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1045] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1046] This invention relates to a system for efficiently searching for images and slides from past documents. This system includes a database for managing past documents, a user interface for inputting search queries, an algorithm for extracting and vectorizing features, and a function for calculating the similarity between the vectors of the documents and the search query.

[1047] System configuration

[1048] 1. Terminal

[1049] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[1050] The terminal is provided with an input interface for the user to input a search query.

[1051] 2. Server

[1052] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[1053] The server receives a search query from a user and vectorizes the query.

[1054] The server calculates the similarity between the vectorized query and the vectors of images or slides stored in the database.

[1055] Based on the results of the similarity calculation, the materials are listed in descending order of similarity.

[1056] 3. Database

[1057] The database stores vectorized data of past images and slides.

[1058] 4. Algorithm

[1059] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used to analyze and vectorize text queries.

[1060] Program processing

[1061] 1. Image / slide acquisition and feature extraction

[1062] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[1063] The server extracts features from these images and slides, analyzing color, shape, texture, and text in the case of diagram images.

[1064] 2. Vectorization

[1065] The server converts the extracted features into vector format, for example, by using a deep learning model to convert image data into a multidimensional vector.

[1066] The extracted vector data is stored in a database.

[1067] 3. Entering a search query and vectorizing it

[1068] The user uses the input interface of the device to enter specific text describing the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[1069] The server receives this text query and vectorizes it using natural language processing techniques, such as the BERT model, to convert the text into multidimensional vectors.

[1070] 4. Similarity calculation

[1071] The server calculates the similarity between the vectorized query and the vectors in the database. It measures the distance using cosine similarity or Euclidean distance, and calculates the similarity.

[1072] 5. Displaying the results

[1073] The server lists images and slides in order of similarity and sends the results to the device.

[1074] The terminal displays the results to the user, who can then check and select the images and slides they need.

[1075] Specific examples

[1076] For example, if a user is creating new planning materials on the theme of "future cities," they may want to reuse an "illustration of a future city" that was included in a previous document. The user enters text that describes the atmosphere of the "future city" into the search interface on their device. The server converts this text into vectors and calculates the similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can then select the "illustration of a future city" from the list and reuse it in the new document. This process allows users to find the materials they need quickly and efficiently.

[1077] The processing flow will be explained below.

[1078] Step 1:

[1079] The device retrieves images and slides from past documents, and loads multiple files containing graphs showing a company's growth prospects and illustrations of future cities.

[1080] Step 2:

[1081] The server extracts the features of the acquired images and slides. Using a deep learning model (e.g., Convolutional Neural Network, CNN), the server analyzes and extracts important features such as the color, shape, and texture of the images, as well as text information within the slides.

[1082] Step 3:

[1083] The server vectorizes the extracted features. The features are expressed as multidimensional vectors, and vectorized data is generated. This vector data is stored in a database.

[1084] Step 4:

[1085] A user uses a search interface on a device to enter a text query that specifies the image or slide atmosphere they want to search for, such as "I want an image that depicts a bright future."

[1086] Step 5:

[1087] The server analyzes and vectorizes the text query received from the user using natural language processing techniques. Specifically, it uses text encoding techniques such as BERT and Word2Vec to convert the text query into a multidimensional vector.

[1088] Step 6:

[1089] The server calculates the similarity between the vectorized query and the vector data of past documents in the database. It measures the distance between each vector using methods such as cosine similarity and Euclidean distance to calculate the similarity.

[1090] Step 7:

[1091] The server lists images and slides in descending order of similarity, generating a list of documents with the highest similarity first, and compiling this list as search results.

[1092] Step 8:

[1093] The server generates a search result list and sends it to the device, containing the images and slides that best match the user's search query.

[1094] Step 9:

[1095] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[1096] In this way, even if a user cannot remember specific keywords, they can efficiently search for and reuse the necessary materials based on the atmosphere of the images and slides.

[1097] Example 1

[1098] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1099] Efficiently searching and retrieving necessary images and slides from past documents is a time-consuming and labor-intensive process. It is especially difficult to quickly find the desired content when a large amount of documents are stored in a database. Furthermore, accurate interpretation of user search queries and provision of optimal documents requires advanced technology.

[1100] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1101] In this invention, the server includes a means for uploading past materials specified by a user, a means for using a deep learning model to extract features of the uploaded materials, and a means for analyzing and vectorizing search queries using a natural language processing model, thereby enabling efficient search for necessary images and slides from past materials and providing the necessary materials with high accuracy.

[1102] "Images and slides" are visual content stored as past documents, and refer to graphics and presentation materials that provide concrete visual information.

[1103] "Features" are data that numerically represent specific properties or patterns extracted from digital data such as images or slides.

[1104] "Vectorization" is the process of representing features as vectors in a multidimensional space, which makes it easier to calculate similarities between data.

[1105] A "search query" is a text input used by a user to search for specific information, thereby providing guidance to the system in locating appropriate materials.

[1106] "Natural language processing technology" refers to a computational technology for analyzing and understanding human language, and in particular to the technology for semantically processing input text.

[1107] A "deep learning model" is a type of machine learning model that uses a multi-layer neural network to automatically learn complex patterns and features from data.

[1108] "Cosine similarity" is a method for calculating the similarity between two vectors, and uses the cosine value of the angle between the vectors.

[1109] "Euclidean distance" is a method for calculating the straight-line distance between two points (vectors). The smaller the distance, the more similar the two points (vectors) are considered to be.

[1110] A "database" is a system for efficiently storing, managing, and searching structured data.

[1111] This invention is a system that efficiently searches for images and slides from past documents and provides them to users. This system is composed of terminals, servers, databases, deep learning models, and natural language processing models.

[1112] First, the device retrieves images and slides from past documents, including hardware such as online storage and local file systems. The user can then specify the documents using the device's input interface and import them into the system.

[1113] The captured images and slides are then processed on a server where features are extracted. The server uses deep learning libraries (e.g., TensorFlow and PyTorch) to analyze image features such as color, shape, texture, and text. These features are represented as multidimensional vectors.

[1114] The server then uses a pre-trained deep learning model (e.g., InceptionV3 or ResNet) to convert these extracted features into vector format, which is then stored in a database (e.g., MongoDB or MySQL).

[1115] When a user enters a search query, they use the input interface of their device. For example, they enter specific text such as "illustration of a futuristic city" and click the search button. This text query is then vectorized by the server using a natural language processing model (e.g., BERT or Word2Vec).

[1116] The vectorized search query is used to calculate the similarity between the vector data of images and slides in the database. The server uses cosine similarity or Euclidean distance for this similarity calculation. Based on the calculated similarity, materials with high similarity are listed.

[1117] Finally, the server sends a list of materials with high similarity to the terminal, which displays it to the user. The user can then select and reuse the images and slides they need from the displayed materials. This system allows users to quickly and efficiently find the materials they need.

[1118] For example, consider a user creating new project materials on the theme of "future cities." The user enters "illustration of a future city" into the search interface on their device and clicks the search button. The server vectorizes this text and calculates its similarity with the vectors in the database. Images and slides with high similarity are listed and displayed on the device. The user can select the "illustration of a future city" from the list and reuse it in the new material. This process allows users to quickly and efficiently find the materials they need.

[1119] Prompt Sentence Examples

[1120] "I want to create a planning document for a future city. Please search past documents for images and slides on the theme of future cities."

[1121] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1122] Step 1: Acquire images and slides

[1123] The device retrieves images and slides from past documents. The user uses a device such as a PC or smartphone to select a specific file from online storage or the local file system and clicks the upload button. In this step, the user can upload a "graph showing a company's growth prospects" or "illustration of a future city," and these files are sent to the system. The input data are the image or slide files, and the output data is a notification that the upload has been completed to the server.

[1124] Step 2: Feature extraction

[1125] The server analyzes images and slides received from the device and extracts features. Deep learning libraries (e.g., TensorFlow and PyTorch) are used for extraction. The image's color distribution, shape, texture, and included text are automatically analyzed. The input data are the uploaded image or slide files, and the output data is a dataset containing the extracted features.

[1126] Step 3: Vectorization

[1127] The server converts the features into multidimensional vectors. A pre-trained deep learning model (e.g., InceptionV3 or ResNet) is used to map the extracted features into a high-dimensional vector space. The input data are the features extracted in step 2, and the output data is data represented as multidimensional vectors.

[1128] Step 4: Saving to the Database

[1129] The server stores the vectorized data in a database (e.g., MongoDB or MySQL). At this time, metadata (e.g., title, date, category, etc.) of the corresponding image or slide is also stored for each vector. The input data is data and metadata expressed as multidimensional vectors, and the output data is vector data stored in the database.

[1130] Step 5: Enter a search query

[1131] The user enters a search query using the search interface on the device. For example, they enter specific keywords such as "illustration of a futuristic city" or "image showing a bright future," and clicks the search button. The input data is the text of the search query, and the output data is a notification that the query has been sent to the server.

[1132] Step 6: Vectorizing the query

[1133] The server vectorizes the search query received from the user using a natural language processing model (e.g., BERT or Word2Vec). It captures the meaning of the query and represents the text as a multidimensional vector. The input data is the search query text, and the output data is the vectorized query.

[1134] Step 7: Similarity calculation

[1135] The server calculates the similarity between the vectorized query and the vector data in the database using algorithms such as cosine similarity or Euclidean distance. The input data are the vectorized query and the vector data in the database, and the output data is the similarity score.

[1136] Step 8: Viewing the results

[1137] The server lists images and slides in descending order of similarity based on the similarity score and sends the list to the terminal. The terminal displays the results to the user, who can then select the images or slides they need from the displayed list. The input data is the similarity list generated by the server, and the output data is the results displayed on the user's terminal.

[1138] (Application example 1)

[1139] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1140] In today's brick-and-mortar stores, it is difficult for consumers to efficiently search for products based on their preferences and past purchase history. It is also difficult for them to instantly understand the vast amount of product information available in stores and make appropriate selections. Therefore, there is a need for a method that allows consumers to find products more quickly and accurately.

[1141] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1142] In this invention, the server includes means for acquiring images and slides from past materials, means for extracting features from the acquired images and slides and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for calculating the similarity between the vectorized search query and the vectors of the past materials and listing materials with high similarity, means for displaying materials with high similarity, means for searching for retail products based on images, means for vectorizing an image taken as the search query and calculating the similarity with product vectors in a database, and means for listing and displaying products with high similarity. This allows a user to instantly find similar products based on images of products they found in a store.

[1143] "Past materials" refers to previously created images, slides, and document data.

[1144] "Features of images and slides" refer to important elements that make up an image, such as color, shape, texture, and text, extracted from the image or slide.

[1145] "Vectorization" is the process of converting features into points in a multidimensional space.

[1146] A "search query" is the text or image information that a user enters into the system.

[1147] "Similarity" is an index that measures the similarity between two vectors, and there are calculation methods such as cosine similarity and Euclidean distance.

[1148] A "database" is a digital warehouse that stores and manages vectorized data extracted from past documents.

[1149] "Natural language processing technology" is a technology that understands and analyzes text data and converts it into a format that can be processed by machines.

[1150] A "deep learning model" is an artificial intelligence technology that automatically learns features using large amounts of data and makes highly accurate predictions and classifications.

[1151] "Retail" is a general term that refers to retail trade and brick-and-mortar sales.

[1152] A "means for calculating similarity" refers to a method or technique for comparing a vectorized query with a vector in a database and measuring the similarity between them.

[1153] This invention relates to a system for improving the shopping experience in brick-and-mortar stores. The system aims to efficiently search for and present related and recommended products to users based on their past purchase history and images and descriptions of products seen in stores.

[1154] The system uses a smartphone as hardware and databases (e.g., MongoDB), deep learning models (e.g., CNN using TensorFlow), and natural language processing models (e.g., BERT) as software. It uses Node.js and Express on the server side and React Native on the front end.

[1155] A user uses the camera function of their smartphone to take a picture of a product they see in a store. This image is sent to the server via the device. The server extracts features from the received image and vectorizes them using a deep learning model. These features include the product's shape, color, texture, etc.

[1156] The server stores the vectorized feature data in a database. Furthermore, the server receives search queries (e.g., "red dress" or "sports shoes") entered by users as text and vectorizes them using natural language processing techniques. The BERT model is used for this vectorization.

[1157] The server calculates the similarity between the vectorized search query and the product vectors in the database. This similarity calculation uses cosine similarity or Euclidean distance. The server then lists the products in descending order of similarity and sends the results to the smartphone.

[1158] The smartphone displays products sorted by similarity, and users can select the product they are interested in and check its details. This allows users to instantly find similar products based on the image of the product they found in the store.

[1159] For example, if a user is looking for a "red dress" in a store, they can take a photo of the product with their smartphone and enter the keyword "red dress." The server analyzes the features based on the image and keyword, and vectorizes them using a deep learning model and natural language processing technology. The server then calculates the similarity with products in the database and displays the results on the user's smartphone.

[1160] Examples of prompts include:

[1161] I'm looking for a "red dress." Show me products similar to this in-store image.

[1162] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1163] Step 1:

[1164] A user takes a photo of a product in a store using the camera on their smartphone.

[1165] Input: Image taken inside the store

[1166] Output: Captured image data

[1167] Image data is sent to the server, which receives the data and begins processing it.

[1168] Step 2:

[1169] The server extracts features from the received image data.

[1170] Input: Photographed image data

[1171] Output: Feature vector

[1172] The server uses a deep learning model (e.g., CNN using TensorFlow) to extract features such as color, shape, and texture from the image.

[1173] Step 3:

[1174] The server vectorizes the extracted features.

[1175] Input: extracted features

[1176] Output: A vector of features

[1177] The server converts these features into multidimensional vectors, which are then stored in a MongoDB database.

[1178] Step 4:

[1179] A user inputs a search query through an input interface of a smartphone.

[1180] Input: Search query (e.g. "red dress")

[1181] Output: Query text

[1182] The terminal transmits this text data to the server.

[1183] Step 5:

[1184] The text data of the search query received by the server is vectorized.

[1185] Input: Text data of the search query

[1186] Output: A vector of text

[1187] The server analyzes this text using natural language processing techniques (e.g., the BERT model) and vectorizes it.

[1188] Step 6:

[1189] The server calculates the similarity between the vectorized search query and the product vectors in the database.

[1190] Input: Search query vector and product feature vector data

[1191] Output: Similarity score

[1192] The server calculates the similarity between the query vector and the product vector using cosine similarity or Euclidean distance.

[1193] Step 7:

[1194] The server lists products in order of similarity and sends the results to the smartphone.

[1195] Input: Similarity score

[1196] Output: List of similar products

[1197] The server lists products with high similarity and sends the results to the terminal.

[1198] Step 8:

[1199] The terminal displays the search results to the user.

[1200] Input: List of similar products

[1201] Output: Displayed product list

[1202] Users can view search results on their smartphones, select products that interest them, and read more details.

[1203] This system allows users to instantly find similar products based on images of products they find in stores.

[1204] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1205] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, the emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[1206] System configuration

[1207] 1. Terminal

[1208] The terminal retrieves previously created planning and proposal materials and sends them to the system.

[1209] The user interface provides a means for users to enter search queries and also includes an interface for accepting facial recognition and voice input from users.

[1210] 2. Server

[1211] The server extracts the features of images and slides obtained from the terminal and vectorizes them.

[1212] The search query received from the user interface is received as text, and the user's emotion recognized from the emotion engine is also received.

[1213] The server vectorizes the search query and corrects it based on sentiment data.

[1214] The similarity between the vectorized query and the vectors of past materials is calculated, and materials with high similarity are listed.

[1215] 3. Database

[1216] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[1217] 4. Algorithm

[1218] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for analyzing text queries.

[1219] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[1220] Program processing

[1221] 1. Image / slide acquisition and feature extraction

[1222] The device retrieves images and slides from past documents, such as graphs showing a company's growth prospects or illustrations of future cities.

[1223] The server extracts features from these images and slides. For images, it analyzes and extracts color, shape, texture, and text information from the slides.

[1224] 2. Vectorization

[1225] The server converts the extracted features into vector format, which allows the image data to be represented as a multidimensional vector.

[1226] These vector data are stored in a database.

[1227] 3. Entering a search query and vectorizing it

[1228] A user uses a device to input a text query describing the image or slide atmosphere they want to search for. For example, they might input the text, "I want an image that depicts a bright future."

[1229] The device sends the user's facial expression and voice data to the emotion engine to recognize the user's emotions. For example, if a user searches for "bright future" with a smile, the emotion engine will recognize this as a "positive emotion."

[1230] The server receives the search query and sentiment data, vectorizes the text query, and corrects it based on sentiment.

[1231] 4. Similarity calculation

[1232] The server calculates the similarity between the vectorized query and the vectors in the database, using methods such as cosine similarity or Euclidean distance.

[1233] 5. Displaying the results

[1234] The server lists images and slides in order of similarity and sends the results to the device.

[1235] The device displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[1236] Specific examples

[1237] For example, if a user is creating a new project document with the theme of "future cities," they may want to reuse an illustration of a future city from a previous document. The user enters text that describes the atmosphere of a "future city" into the device's search interface and intentionally smiles. The device recognizes the facial expression and sends it to the server as positive emotion data. The server receives the text query and emotion data and generates appropriate search results based on them. This allows the illustration the user is looking for to be efficiently reused from previous documents. This process allows users to find the materials they need quickly and efficiently.

[1238] The processing flow will be explained below.

[1239] Step 1:

[1240] The terminal retrieves images and slides from past documents. Specifically, it accesses the database or file system within the system and reads images and slides contained in planning documents and proposal documents.

[1241] Step 2:

[1242] The server extracts features from the acquired images and slides. The server uses a deep learning model (e.g., Convolutional Neural Network, CNN) to analyze the color, shape, and texture of the images, as well as the text information in the slides, and extracts features.

[1243] Step 3:

[1244] The server vectorizes the extracted features, expresses the features as multidimensional vectors, and generates vectorized data. These vector data are stored in a database.

[1245] Step 4:

[1246] A user uses the device's search interface to enter a text query that specifies the image or slide mood they are looking for, for example, "I want images that depict a bright future."

[1247] Step 5:

[1248] The emotion engine receives user input and analyzes the user's facial expression and voice data to recognize emotions. Specifically, it captures the user's facial expressions through a camera and identifies emotions using a facial expression analysis algorithm. It also captures voice data through a microphone and analyzes emotions using a voice emotion recognition algorithm.

[1249] Step 6:

[1250] The server receives the user's text query and the recognized emotion data from the emotion engine, vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec), and corrects the search query based on the recognized emotion.

[1251] Step 7:

[1252] The server calculates the similarity between the vectorized query and the data vectors in the database, using methods such as cosine similarity or Euclidean distance to measure the distance between each vector and calculate the similarity.

[1253] Step 8:

[1254] The server lists images and slides in order of similarity, compiles the results as search results, generates a search result list, and sends the results to the device.

[1255] Step 9:

[1256] The terminal displays the search results to the user, who can then select images or slides that fit their purpose and reuse them in creating materials.

[1257] Specific examples

[1258] For example, when creating new planning materials with the theme of "future cities," a user might want to reuse an "illustration of a future city." The user enters text describing the atmosphere of the "future city" into the device's search interface and then captures their own smile with the camera. The emotion engine recognizes the user's smile and generates emotional data representing a "positive emotion." The server receives the text query and the emotional data, and corrects the search query based on both to generate optimal search results. As a result, the "illustration of a future city" desired by the user is efficiently listed and can be reused in creating new planning materials. This process, with the help of the emotion engine, enables searches that more accurately reflect the user's intent.

[1259] Example 2

[1260] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1261] Conventional document search systems not only have difficulty efficiently finding the necessary images and slides from a large number of digital files, but also fail to consider the user's emotions, resulting in search results that often deviate from the user's intentions. Furthermore, feature extraction and search query vectorization rely too heavily on user input, resulting in poor search result accuracy.

[1262] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1263] In this invention, the server includes means for acquiring digital files from past materials, means for extracting features from the acquired digital files and vectorizing them, means for receiving a search query entered by a user as text and vectorizing it, means for recognizing the user's emotions and generating emotion data, means for correcting the vectorized search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past materials and listing materials with high similarity, and means for displaying materials with high similarity. This makes it possible to provide highly accurate search results while taking user emotions into consideration.

[1264] "Past materials" are digital files that were created or collected in the past and registered in the system, and primarily refer to images and slides.

[1265] "Digital files" refers to data stored and exchanged electronically, and generally includes image files and slide files.

[1266] A "feature" is an attribute value or pattern extracted from a digital file, and is information that allows the characteristics of the digital file to be quantified or vectorized.

[1267] "Vectorization" is a technique that represents digital files or text queries as multidimensional numerical vectors, making them easier to process on a computer.

[1268] "Search query" refers to a text input by a user that indicates to the system what they wish to search for.

[1269] "Emotion data" refers to data that expresses the emotional state recognized from the user's facial expressions and voice as numerical values ​​or categories.

[1270] "Similarity" is an index that indicates the closeness or degree of agreement between two vectors (a search query and a digital file's features), and is calculated using methods such as cosine similarity or Euclidean distance.

[1271] "Listing" refers to displaying a list of digital files obtained as search results in order of similarity.

[1272] This invention is a system for efficiently searching for images and slides from past documents, equipped with a function for recognizing user emotions. This system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion recognition engine, and a function for calculating the similarity between vectors. The hardware and software used are described below.

[1273] Hardware and Software Configuration

[1274] 1. Terminal

[1275] Devices used as terminals include PCs, tablets, smartphones, etc.

[1276] The device provides a user interface (UI) for users to enter search queries, which includes not only text input but also facial expression recognition and voice input functions.

[1277] 2. Server

[1278] The server is the main component that processes the data sent from the terminals.

[1279] Deep learning models (e.g., Convolutional Neural Network, CNN) are used to extract features from images and slides.

[1280] Natural language processing techniques (e.g., BERT, Word2Vec) are used to parse and vectorize text queries.

[1281] The emotion engine uses technology to recognize the user's facial expressions and identify emotions from their voice (e.g., facial expression analysis, voice emotion recognition).

[1282] 3. Database

[1283] The features of past images and slides are stored in vector format, and the database also includes an index that allows fast and efficient search of vector data.

[1284] Specific operation of the system

[1285] Suppose a user searches for "an illustration that expresses the atmosphere of a futuristic city." The user enters a text query through the device's search interface and intentionally smiles. The device recognizes this facial expression in real time and transmits it to the server as positive emotion data.

[1286] The server vectorizes the text query and the received emotion data, and then modifies the query based on the emotion data. The server then compares the vectors with those of past digital files to identify those with high similarity, using, for example, cosine similarity or Euclidean distance, to generate optimal search results.

[1287] Finally, the server sends the search results to the terminal, which displays them to the user. The user can select images or slides that suit their purpose from the search results and reuse them in new materials. This allows users to quickly and accurately find the materials they need.

[1288] Example prompt sentence:

[1289] Search for "I want an illustration that conveys the atmosphere of a futuristic city."

[1290] Please implement a system that allows users to search for "future cities" with a smile.

[1291] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1292] Step 1: User enters a search query

[1293] The user uses the device to input the desired image or slide content as text into the search interface. For example, they might input "I'm looking for an illustration that evokes the atmosphere of a futuristic city." At this point, the search query text is generated as input data. Once the user has finished entering the query, the device passes this input data to the next processing step.

[1294] Step 2: Recognizing user emotion data by the device

[1295] As the user enters a text query, the device's camera and microphone capture the user's facial expressions and voice. The device then uses an emotion recognition engine to analyze this data and recognize the user's emotion. The emotion data generated is then sent to the next processing step.

[1296] Step 3: Obtaining past documents via terminal

[1297] The device retrieves previously created materials (digital files) from the company's cloud storage or internal server. The input here is past materials retrieved according to conditions specified by the user. The retrieved past materials (images and slides) are sent to the next processing step.

[1298] Step 4: Feature extraction and vectorization on the server

[1299] The server extracts features from images and slides of past documents received from the terminal. A deep learning model (e.g., CNN) is used to extract features. At this time, the input data is past documents that meet the user's requirements, and the output data is the features extracted from these documents. The features are converted into vector format and stored in a database.

[1300] Step 5: Vectorizing search queries by the server

[1301] The server vectorizes the text query entered by the user using natural language processing techniques (e.g., BERT, Word2Vec). Here, the input is the text query and the output is the vectorized query. This vectorized query is sent to the next processing step.

[1302] Step 6: Server-based query refinement based on emotion data

[1303] The server uses the sentiment data to correct the vectorized query. For example, if there is positive sentiment data, it adds a positive weight to the corresponding search query vector. Here, the input is the vectorized query and sentiment data, and the output is the corrected vector query.

[1304] Step 7: Similarity calculation by the server

[1305] The server calculates the similarity between the corrected vector query and the document vectors in the database. The algorithms used include cosine similarity and Euclidean distance. The input data are the corrected query vector and the document vector, and the output is a list of documents with high similarity.

[1306] Step 8: Server lists and sends search results

[1307] As a result of the similarity calculation, the server lists the materials in descending order of similarity and sends it to the terminal. The input here is the result of the similarity calculation, and the output is a list of search results. The list of listed materials is sent to the terminal.

[1308] Step 9: Device displays search results and user selection

[1309] The terminal displays the received search results on the user interface. The user can select images or slides that suit their purpose from this list. The input here is the list of search results, and the output is the materials selected by the user. The user can reuse the selected materials in new project materials to create.

[1310] (Application example 2)

[1311] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1312] When searching for images or slides from past documents, it is difficult to obtain appropriate results that reflect the user's emotions. In particular, if search results are generated only by simple text matching or vector comparison without considering the user's emotions when searching, the optimal results may not be provided according to the user's expectations.

[1313] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1314] In this invention, the server includes means for extracting and vectorizing acquired images and slides as features, means for vectorizing search queries entered by users, means for recognizing user emotions and correcting the search query based on the emotion data, means for calculating the similarity between the vectorized search query and the vectors of past documents and listing documents with high similarity, and means for receiving user emotions as feedback and reflecting them in the next search, thereby making it possible to provide optimal search results that reflect the user's emotions.

[1315] - "Historical materials" refers to data such as digital documents, images, and slides that have been collected and stored to date.

[1316] "Image" refers to any visual data displayed or stored by a digital device.

[1317] "Slides" refers to individual visual pages used in presentations and reports.

[1318] "Means of acquisition" refers to a method or device for capturing data of historical materials from a designated location.

[1319] "Means for extracting features" refers to a method or device that analyzes the visual and content elements of an image or slide and converts them into numerical data.

[1320] "Vectorization means" refers to a method or device that formalizes extracted features as multidimensional vectors.

[1321] "Search query" refers to a string of characters or spoken instructions entered by a user to indicate the information or material they are seeking.

[1322] "Means for recognizing emotions" refers to technology or devices that determine a user's psychological state at that time from facial expressions and voice data.

[1323] "Means for correcting based on emotional data" refers to technologies or devices that adjust or transform search queries taking into account the emotional state of the user.

[1324] The term "means for calculating similarity" refers to an algorithm or device for numerically evaluating the similarity between vector-format data.

[1325] The "listing means" refers to a method or device for preferentially extracting data with high similarity and displaying it in a list format.

[1326] "Displaying means" refers to a method or device for visually presenting search results to a user.

[1327] "Means for receiving feedback" refers to a method or device for incorporating user evaluations and opinions into the system.

[1328] A "deep learning model" refers to an algorithm that uses an advanced neural network to extract and learn features.

[1329] "Natural language processing technology" refers to all technologies for analyzing and understanding natural language texts that humans use on a daily basis.

[1330] This invention combines a system for efficiently searching for images and slides from past documents with an emotion engine that recognizes user emotions. The system includes a database for managing past documents, a user interface for entering search queries, an algorithm for extracting and vectorizing features, an emotion engine, and a function for calculating the similarity between the search query and the document vectors.

[1331] System configuration

[1332] 1. Terminal

[1333] The terminal retrieves previously created planning and proposal materials and sends them to the system. The user interface provides a means for users to input search queries. It also includes an interface that recognizes the user's facial expressions and receives voice input. Specifically, a smartphone or smart glasses can function as the terminal.

[1334] 2. Server

[1335] The server extracts features from images and slides obtained from the device and vectorizes them. A deep learning model (e.g., CNN) is used to extract features, and natural language processing technology (e.g., BERT, Word2Vec) is used to analyze text queries. The server receives the search query as text from the user interface, as well as the user's emotions recognized by the emotion engine. The server vectorizes the search query and corrects it based on the emotion data. The server calculates the similarity between the vectorized query and the vectors of past materials and lists materials with high similarity.

[1336] 3. Database

[1337] The feature values ​​of past images and slides are stored as vectors, which contain multiple images and slides.

[1338] 4. Algorithm

[1339] Deep learning models (e.g., CNN) are used for feature extraction, and natural language processing techniques (e.g., BERT, Word2Vec) are used for text query analysis. The emotion engine uses technologies to recognize user facial expressions and identify emotions from voice (e.g., facial expression analysis, voice emotion recognition).

[1340] Program processing

[1341] The server uses a deep learning model to extract features from images and slides sent from the device and vectorizes them. The device receives the search query entered by the user through the user interface as text and recognizes the user's emotions using an emotion engine. The server receives this text query and emotion data, vectorizes the query using natural language processing technology, and corrects it based on the emotion data. The server then calculates the similarity between the vectorized query and the vectors of past documents and lists documents with high similarity. Finally, the search results are displayed to the user via the device. The user can select images or slides that match their purpose from the displayed results and reuse them in creating documents.

[1342] Specific examples

[1343] For example, imagine a user is searching for "new sneakers" in a physical store. The user launches a smartphone app, enters "comfortable sneakers" in the search field, and intentionally smiles. The app recognizes this facial expression and sends it to the server as a positive emotion.

[1344] Example prompt (input to generative AI model):

[1345] A user is searching for "comfortable sneakers." The emotion engine recognizes the user's smile and confirms that the emotion is positive. Find the most suitable sneakers based on past purchase history and product reviews. Also, display related product recommendations in the search results.

[1346] This process allows users to quickly find the product that best suits their emotions, improving the shopping experience in physical stores. Specifically, it lists comfortable sneakers and related products that users have previously purchased, and allows users to try them on and provide feedback to further improve the service.

[1347] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1348] Step 1:

[1349] The terminal retrieves images and slides from past documents and sends this data to the system. Specifically, it retrieves previously created planning and proposal documents from a database. The input is past document data, and the output is data sent to the server.

[1350] Step 2:

[1351] The server uses a deep learning model (e.g., CNN) to extract features from acquired images and slides and vectorize them. Specifically, it analyzes and extracts image color, shape, texture, and text information from slides. The input is image or slide data, and the output is feature data in vector format.

[1352] Step 3:

[1353] The user uses the device to input the mood of the image or slide they want to search for as a text query. At this time, the device also sends the user's facial expressions and voice input to the emotion engine to recognize emotions. The input is the user's text query and emotion data, and the output is data sent from the device to the server.

[1354] Step 4:

[1355] The server analyzes and vectorizes the text query using natural language processing techniques (e.g., BERT, Word2Vec). It then corrects the vector based on emotion data from the emotion engine. The input is the text query and emotion data, and the output is the corrected vector-formatted search query.

[1356] Step 5:

[1357] The server calculates the similarity between the vectorized search query and the vectors of past documents using methods such as cosine similarity or Euclidean distance. The input is the corrected search query vector and the vectors of past documents, and the output is a list of documents with high similarity.

[1358] Step 6:

[1359] The server lists the search results in order of similarity and sends them to the terminal. The input is a list of similarities between documents, and the output is the transmission of the search results to the terminal.

[1360] Step 7:

[1361] The terminal displays the search results to the user. The user selects images and slides that fit the purpose from the displayed list and reuses them in creating materials. The input is the search results sent from the server, and the output is a visual presentation to the user.

[1362] Step 8:

[1363] The user inputs feedback such as satisfaction level for the selected result, and the terminal transmits this feedback to the server. The input is the user's feedback data, and the output is data transmission to the server.

[1364] Step 9:

[1365] The server reflects the received feedback in the emotion engine to improve the accuracy of the next search results. The input is the feedback data, and the output is an improved emotion model.

[1366] 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 control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[1367] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1368] 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 the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[1369] The emotion identification model 59 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 an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1370] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[1371] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[1372] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[1373] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[1374] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs 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 a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[1375] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[1376] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[1377] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[1378] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[1379] 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.

[1380] It is not necessary to store all 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 all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[1381] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[1382] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with 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). Also, the hardware resource that executes the specific processing may be a single processor.

[1383] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[1384] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[1385] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[1386] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[1387] The following is further disclosed regarding the above embodiment.

[1388] (Claim 1)

[1389] A means of retrieving images and slides from past documents;

[1390] A means for extracting and vectorizing the acquired images and slides as features;

[1391] A means for receiving a search query entered by a user as text and vectorizing it;

[1392] A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity;

[1393] a means for displaying highly similar materials;

[1394] A system including:

[1395] (Claim 2)

[1396] 10. The system of claim 1, wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

[1397] (Claim 3)

[1398] The system of claim 1, wherein features of images or slides are extracted using a deep learning model.

[1399] "Example 1"

[1400] (Claim 1)

[1401] A means of retrieving images and slides from past documents;

[1402] A means for extracting and vectorizing the acquired images and slides as features;

[1403] A means for receiving a search query entered by a user as text and vectorizing it;

[1404] A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity;

[1405] a means for displaying highly similar materials;

[1406] a means for uploading user-specified past materials;

[1407] a means for using a deep learning model to extract features of the uploaded material;

[1408] A means for analyzing and vectorizing search queries using a natural language processing model;

[1409] A means for using cosine similarity or Euclidean distance for similarity calculation;

[1410] A system including:

[1411] (Claim 2)

[1412] 10. The system of claim 1, wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

[1413] (Claim 3)

[1414] The system of claim 1, wherein features of images or slides are extracted using a deep learning model.

[1415] "Application Example 1"

[1416] (Claim 1)

[1417] A means of retrieving images and slides from past documents;

[1418] A means for extracting and vectorizing the acquired images and slides as features;

[1419] A means for receiving a search query entered by a user as text and vectorizing it;

[1420] A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity;

[1421] a means for displaying highly similar materials;

[1422] a means for searching retail items based on an image;

[1423] A means for vectorizing an image taken as a search query and calculating the similarity between the image and a product vector in a database;

[1424] A means of listing and displaying highly similar products;

[1425] A system including:

[1426] (Claim 2)

[1427] 10. The system of claim 1, wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

[1428] (Claim 3)

[1429] The system of claim 1, wherein features of images or slides are extracted using a deep learning model.

[1430] "Example 2: Combining Emotion Engines"

[1431] (Claim 1)

[1432] A means of obtaining digital files from historical materials;

[1433] A means for extracting and vectorizing the acquired digital file as features;

[1434] A means for receiving a search query entered by a user as text and vectorizing it;

[1435] means for recognizing a user's emotion and generating emotion data;

[1436] a means for correcting the vectorized search query based on sentiment data;

[1437] A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity;

[1438] a means for displaying highly similar materials;

[1439] A system including:

[1440] (Claim 2)

[1441] 10. The system of claim 1, wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

[1442] (Claim 3)

[1443] The system of claim 1, wherein features of a digital file are extracted using a machine learning model.

[1444] "Application example 2 when combining emotion engines"

[1445] (Claim 1)

[1446] A means of retrieving images and slides from past documents;

[1447] A means for extracting and vectorizing the acquired images and slides as features;

[1448] A means for receiving a search query entered by a user as text and vectorizing it;

[1449] A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity;

[1450] a means for displaying highly similar materials;

[1451] means for recognizing user emotions and amending search queries based on the emotion data;

[1452] A way to receive user feedback and reflect it in the next search.

[1453] A system including:

[1454] (Claim 2)

[1455] 10. The system of claim 1, wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

[1456] (Claim 3)

[1457] The system of claim 1, wherein features of images or slides are extracted using a deep learning model. [Explanation of symbols]

[1458] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. A means of retrieving images and slides from past documents; A means for extracting and vectorizing the acquired images and slides as features; A means for receiving a search query entered by a user as text and vectorizing it; A method for calculating the similarity between the vectorized search query and the vectors of past documents, and listing documents with high similarity; a means for displaying highly similar materials; A system including:

2. The system of claim 1 , wherein the text of the search query is analyzed and vectorized using natural language processing techniques.

3. The system of claim 1 , wherein features of images and slides are extracted using a deep learning model.

Images