Information provision apparatus and information provision method based on generative artificial intelligence model

Abstract

An information provision apparatus based on a generative artificial intelligence model, according to an embodiment of the present invention, may comprise a processor that: receives a prompt related to a product from at least one user terminal; identifies first metadata corresponding to the prompt from among metadata corresponding to product information divided into a plurality of chunks on the basis of a document structure; generates result information corresponding to the prompt on the basis of the first metadata and a first plurality of chunks corresponding to the first metadata; and transmits the result information to the user terminal.

Description

Information providing device and information providing method based on generative artificial intelligence model

[0001] The present invention relates to an information providing device and an information providing method based on a generative artificial intelligence model.

[0002] Recently, with the development of related technologies such as the Internet of Things (IoT) and Artificial Intelligence (AI), smart factories have become a hot topic. A smart factory is a general term for a manufacturing system operated by applying Information and Communications Technology (ICT) combined with digital automation solutions to production processes such as design, development, and manufacturing.

[0003] One of the key requirements in operating a smart factory is providing information that is closely related to the products or the company and possesses high importance and practicality. To this end, it is necessary to devise measures that enable stakeholders, including equipment managers, to quickly and accurately find the information they need.

[0004] Meanwhile, Generative Artificial Intelligence (GAI) is a type of artificial intelligence technology that generates or creates new data based on given input data. Chatbots, which are an example of applications for Generative AI models, are systems that provide necessary information through conversation with users and are utilized in various fields for purposes such as user support, information provision, and task automation. However, problems have been raised that the accuracy and usefulness of generated content cannot be guaranteed, as it is difficult to accurately understand and interpret all user inputs using only simple Natural Language Processing (NLP) technology, and a so-called hallucination phenomenon occurs where incorrect information is provided regarding unlearned content.

[0005] To address this, a new type of architecture called Retrieval-Augmented Generation (RAG) was developed to improve text generation performance by integrating the Large Language Model (LLM) of generative AI models with information retrieval models.

[0006] The objective of the present invention is to provide an information providing device and an information providing method based on a generative artificial intelligence model that provides information about a product more quickly and accurately.

[0007] In an information providing device based on a generative artificial intelligence model according to an embodiment of the present invention, the device may include a processor that receives a prompt regarding a product from at least one user terminal, identifies a first metadata corresponding to the prompt among metadata corresponding to product information divided into a plurality of chunks based on a document structure, generates result information corresponding to the prompt based on the first metadata and a first plurality of chunks corresponding to the first metadata, and transmits the result information to the user terminal.

[0008] The processor can generate the result information based on the similarity between the embedding vectors of the prompt and the first metadata and the embedding vectors of a plurality of chunks and metadata within the vector database using a generative artificial intelligence model.

[0009] The above processor can divide the product information into multiple chunks based on the document structure.

[0010] The processor can identify header information within the product information based on text formatting and map content information placed between the header information to divide the product information into multiple chunks.

[0011] The processor may re-rank the first plurality of chunks based on the similarity between the first plurality of chunks and the prompt, and generate result information based on the re-ranked first plurality of chunks.

[0012] A method for providing information based on a generative artificial intelligence model according to an embodiment of the present invention may include: receiving a prompt regarding a product from at least one user terminal; identifying a first metadata corresponding to the prompt among metadata corresponding to product information divided into a plurality of chunks based on a document structure; generating result information corresponding to the prompt based on the first metadata and a first plurality of chunks corresponding to the first metadata; and transmitting the result information to the user terminal.

[0013] The step of generating the above result information may include the step of generating the above result information based on the similarity between the embedding vector of the prompt and the first metadata and the embedding vector of a plurality of chunks and metadata within the vector database using a generative artificial intelligence model.

[0014] The above information provision method may further include the step of dividing the product information into multiple chunks based on the document structure.

[0015] The step of dividing into a plurality of chunks may include: identifying header information within the product information based on text formatting; and dividing the product information into a plurality of chunks by mapping the header information and the content information placed between the header information.

[0016] The step of generating the result information may include: a step of re-ranking the first plurality of chunks based on the similarity between the first plurality of chunks and the prompt; and a step of generating the result information based on the re-ranked first plurality of chunks.

[0017] According to one embodiment of the present invention, the performance of a generative artificial intelligence model can be enhanced by using chunks that are divided by reflecting contextual information.

[0018] According to one embodiment of the present invention, answer accuracy can be further improved through re-ranking among candidate answers.

[0019] According to one embodiment of the present invention, it is possible to clearly identify the user's intent within a short period of time to provide a more suitable response to the requirements, and faster task processing by relevant personnel and cost reduction effects regarding the operation of product-related consultation centers are expected.

[0020] According to one embodiment of the present invention, customer satisfaction can be improved as the customer can quickly check the information they want to check without having to search for it one by one on the product information.

[0021] FIG. 1 is a schematic diagram illustrating an information providing system according to one embodiment of the present invention.

[0022] FIG. 2 is a block diagram illustrating the configuration of an information providing device according to one embodiment of the present invention.

[0023] FIG. 3 is a diagram illustrating the operation flowchart of an information providing device according to one embodiment of the present invention.

[0024] FIG. 4 is a drawing illustrating the operation of an information providing device according to the first embodiment of the present invention.

[0025] FIG. 5 is a diagram illustrating the operation of an information providing device according to a second embodiment of the present invention.

[0026] FIG. 6 is a diagram illustrating the operation of an information providing device according to an embodiment of the present invention.

[0027] Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. The detailed description disclosed below, together with the accompanying drawings, is intended to describe exemplary embodiments of the present invention and is not intended to represent the only embodiment in which the present invention can be practiced. In order to clearly explain the present invention in the drawings, parts unrelated to the description may be omitted, and the same reference numerals may be used for identical or similar components throughout the specification.

[0028] FIG. 1 is a schematic diagram illustrating an information providing system according to one embodiment of the present invention.

[0029] An information providing system (1) (hereinafter referred to as the system (1)) according to one embodiment of the present invention may include a database (10), an information providing device (100), and a user terminal (200).

[0030] The database (10) is a device that stores information necessary during the process of the information providing device (100) providing information. The database (10) may be implemented by dividing it into two or more databases as needed, and the method of implementation is not limited to any one, such as being implemented as a separate device from the information providing device (100) or being implemented within the information providing device (100).

[0031] The information providing device (100) is a device that provides information in response to a search request from a user terminal (200) and can be implemented as a computer, server, smartphone, tablet PC, smart pad, laptop, etc. The information providing device (100) can provide information about a product through a separately implemented interface, such as a product search application or a web page.

[0032] The user terminal (200) is a device that requests a search for product information from the information providing device (100) using a chatbot program and receives result information from the information providing device (100), and can be implemented as a mobile terminal such as a smartphone, tablet PC, smart pad, or laptop.

[0033] In the present invention, an information provision system (1) is proposed that can supply necessary information more quickly and accurately to personnel, including facility managers of a smart factory.

[0034] Hereinafter, the configuration and operation of an information providing device (100) according to an embodiment of the present invention will be described in detail with reference to the drawings.

[0035] FIG. 2 is a block diagram illustrating the configuration of an information providing device according to one embodiment of the present invention.

[0036] An information providing device (100) according to one embodiment of the present invention may include an input unit (110), a communication unit (120), a display unit (130), a storage unit (140), and a processor (150).

[0037] The input unit (110) generates input data in response to user input of the information providing device (100). For example, the user input may be a user input that initiates the operation of the information providing device (100), a user input required to build a generative artificial intelligence model / database, etc., and may also be applied without limitation if it is a user input required to transmit result information to a user terminal (200).

[0038] The input unit (110) includes at least one input means. The input unit (110) may include a keyboard, a key pad, a dome switch, a touch panel, a touch key, a mouse, a menu button, etc.

[0039] The communication unit (120) can communicate with external devices such as a user terminal (200) and a server to transmit and receive prompts, multiple chunks, metadata, product information, result information, generative artificial intelligence models, etc.

[0040] To this end, the communication unit (120) can perform wireless communication such as 5G (5th generation communication), LTE-A (Long Term Evolution-Advanced), LTE (Long Term Evolution), Wi-Fi (Wireless Fidelity), Bluetooth, or wired communication such as LAN (Local Area Network), WAN (Wide Area Network), and power line communication.

[0041] The display unit (130) displays display data according to the operation of the information providing device (100). The display unit (130) can display a screen that displays result information, a screen that receives user input, etc.

[0042] The display unit (130) includes a Liquid Crystal Display (LCD), a Light Emitting Diode (LED) display, an Organic LED (OLED) display, a Micro Electro Mechanical Systems (MEMS) display, and an electronic paper display. The display unit (130) can be combined with the input unit (110) to be implemented as a touch screen.

[0043] The storage unit (140) stores operation programs of the information providing device (100). The storage unit (140) includes storage with non-volatile properties that can preserve data (information) regardless of whether power is provided, and memory with volatile properties in which data to be processed by the processor (150) is loaded and data cannot be preserved if power is not provided. Storage includes flash memory, HDD (Hard-Disc Drive), SSD (Solid-State Drive), ROM (Read Only Memory), etc., and memory includes buffer, RAM (Random Access Memory, RAM), etc.

[0044] As described above, the storage unit (140) may be implemented by including any one of the databases (10). The storage unit (140) may store prompts, multiple chunks, metadata, product information, result information, generative artificial intelligence models, etc., and may store computation programs, etc., required in the process of collecting and searching product information, identifying metadata, vector embedding, building and running generative artificial intelligence models, obtaining result information, etc.

[0045] The processor (150) can execute software, such as a program, to control at least one other component (e.g., hardware or software component) of the information providing device (100) and can perform various data processing or operations.

[0046] A processor (150) according to one embodiment of the present invention receives a prompt regarding a product from at least one user terminal, identifies a first metadata corresponding to the prompt among metadata corresponding to product information divided into a plurality of chunks based on a document structure, generates result information corresponding to the prompt based on the first metadata and a first plurality of chunks corresponding to the first metadata, and can transmit the result information to the user terminal.

[0047] At this time, the processor (150) may build a database and / or generative artificial intelligence model, or receive and store a previously built database and / or generative artificial intelligence model from an external source and use it, and is not limited to either one.

[0048] Meanwhile, the processor (150) may perform at least some of the data analysis, processing, and result information generation for performing the above operations using at least one of machine learning, neural network, or deep learning algorithms as a rule-based or artificial intelligence algorithm. Examples of neural networks may include models such as CNN (Convolutional Neural Network), DNN (Deep Neural Network), RNN (Recurrent Neural Network), and transformer.

[0049] FIG. 3 is a diagram illustrating the operation flowchart of an information providing device according to one embodiment of the present invention.

[0050] The following operations may be performed using a generative artificial intelligence model. The generative artificial intelligence model may be a language model, and its type is not limited to any one, such as a massive language model or a lightweight language model.

[0051] According to one embodiment of the present invention, the processor (150) can receive a prompt regarding a product from at least one user terminal (200) (S10).

[0052] A prompt is a query input into a generative artificial intelligence model, and any inquiry related to a product is sufficient. For example, it may be the rated conditions of a specific product, maintenance methods for the product, etc. The prompt may be voice input in addition to text input, and the prompt format is not limited to any one. The processor (150) can receive prompts regarding the product through a chatbot program.

[0053] According to one embodiment of the present invention, the processor (150) can identify a first metadata corresponding to a prompt among metadata corresponding to product information divided into a plurality of chunks based on a document structure (S20).

[0054] Product information includes all information regarding the product and may include user manuals, product guidelines, maintenance manuals, test reports, data sheets, certificates, Digital Product Passport (DPP), catalogs, test data, etc.

[0055] In addition to the above, product information may include product introduction videos, drawings (circuit diagrams, wiring diagrams, exploded perspective views, etc.), user manual videos, product guideline videos, accessory replacement manual videos, etc. As such, product information is not limited by type or format and includes audio files, video files, image files, etc., in addition to documents.

[0056] However, product information subject to chunking is based on text-based information, and other product information consisting solely of images, video, or audio files is linked to the relevant chunk as metadata.

[0057] Chunking refers to dividing product information into units (called chunks) that are easy for a generative artificial intelligence model to learn. A processor (150) can divide product information into multiple chunks based on the document structure. When product information is divided according to the document structure, the logical flow and semantic units of the product information are maintained, making it easy to grasp the context of the entire document even if each chunk is read independently. In addition, since the chunks are aligned with a logical structure, the same topics or keywords are grouped together, making searching easier. Through this, not only the speed of the answer but also the accuracy of the answer can be increased when summarizing document content or conducting Q&A. An example of chunking is illustrated in FIG. 5.

[0058] The chunking method can be varied and is not limited to any one. For example, the processor (150) can identify header information within product information based on text formatting. Text formatting may include font, font size, specific patterns (numbers, uppercase letters, symbols), line breaks, etc. Header information is information representing the content included in each chunk and may be a title (main title, subtitle, etc.). The processor (150) can identify header information using a separate program (e.g., Markdown, etc.).

[0059] The processor (150) can map header information and content information to divide product information into multiple chunks. Content information is content placed between header information, and its format is not limited to any one of the following: text, table, drawing, etc.

[0060] Meanwhile, when a table is included in a chunk, the processor (150) can recognize text included in the table using character recognition technology. The character recognition technology may be, for example, Optical Character Recognition (OCR) technology, but is not limited thereto.

[0061] Metadata is information used to quickly search for chunks associated with a prompt and can be generated for each chunk. Metadata may include the chunk creation date and time, chunk storage location, title (part of header information), and information linking to information associated with the chunk (e.g., images, drawings) (links, storage locations, etc.); additionally, any information describing each chunk may be included without restriction.

[0062] The processor (150) can identify first metadata corresponding to a prompt among the metadata. For example, the processor (150) can identify first metadata containing the keyword among the metadata by extracting the keyword of the prompt. As another example, the processor (150) can identify the first metadata based on the similarity between the embedding vector of the prompt and the embedding vector of the metadata. This is a process for filtering chunks associated with the answer among countless chunks (also called document filtering), which can lead to fast and accurate searching and reduce the amount of computation.

[0063] According to one embodiment of the present invention, the processor (150) can generate result information corresponding to a prompt based on first metadata and a first plurality of chunks corresponding to the first metadata (S30).

[0064] The processor (150) can generate result information based on the similarity between the embedding vectors of the prompt and the first metadata and the embedding vectors of the multiple chunks and metadata within the vector database using a generative artificial intelligence model. Similarity can be determined by distance measurement methods such as Euclidean distance or cosine similarity, and similarity can also be measured in various other known ways. The processor (150) can identify a first multiple chunk corresponding to the first metadata among the multiple chunks. Specific details regarding this are described with reference to FIGS. 4 and FIGS.

[0065] According to one embodiment of the present invention, the processor (150) can transmit result information to the user terminal (200) (S40).

[0066] The processor (150) can display result information on the user terminal (200) through a chatbot program. The result information is detailed information about a product searched through a prompt. For example, if the prompt requests the product's rated conditions, the content of a chunk related to the product's rated conditions can be summarized and provided as result information. As another example, if the prompt requests a drawing of the product, the drawing can be provided as result information.

[0067] Meanwhile, the processor (150) can provide multiple result information to the user terminal (200) and can receive user input from the user terminal (200) to select one of the result information.

[0068] According to one embodiment of the present invention, it is possible to clearly identify the user's intent within a short period of time to provide a more suitable response to the requirements, and faster task processing by relevant personnel and cost reduction effects regarding the operation of product-related consultation centers are expected.

[0069] According to one embodiment of the present invention, customer satisfaction can be improved as the customer can quickly check the information they want to check without having to search for it one by one on the product information.

[0070] FIG. 4 is a drawing illustrating the operation of an information providing device according to the first embodiment of the present invention.

[0071] Figure 4 borrows the content described above with reference to Figure 3, and specific explanations of overlapping content are omitted.

[0072] First, the information providing device (100) can collect product information and chunk the product information. The information providing device (100) can construct a vector database by vector embedding multiple chunks.

[0073] The information providing device (100) can identify a first plurality of chunks related to a prompt using a prompt received from a user terminal (200) and a vector database. For example, the information providing device (100) can identify chunks related to a prompt using the similarity between the embedding vector of the prompt received from the user terminal (200) and the embedding vector of the vector database. Similarity can be determined by distance measurement methods such as Euclidean distance or cosine similarity, and similarity can also be measured in various other known ways.

[0074] After identifying a first plurality of chunks, the information providing device (100) can re-rank the chunks based on the similarity between the first plurality of chunks and the prompt. Re-ranking is a process of prioritizing the chunks that are candidates for the answer, which can help the generative artificial intelligence model generate result information more easily. At this time, the similarity may be similarity between embedding vectors, but there may be various methods of measuring similarity, such as measuring based on whether common keywords are included.

[0075] The information providing device (100) can generate result information based on a first plurality of reranked chunks. The information providing device (100) can generate result information by forming an enhanced prompt including a prompt and the first plurality of reranked chunks, and inputting the enhanced prompt into a generative artificial intelligence model.

[0076] According to one embodiment of the present invention, answer accuracy can be further improved through re-ranking among candidate answers.

[0077] FIG. 5 is a diagram illustrating the operation of an information providing device according to a second embodiment of the present invention.

[0078] Figure 5 describes the chunking of product information as described in relation to S20 of Figure 3.

[0079] FIG. 5 illustrates product information #1 (510) divided into multiple chunks (511, 512, 513) based on a conventional method of semantic chunkers, and product information #2 (520) divided into multiple chunks (521, 522, 523) based on a document structure according to an embodiment of the present invention.

[0080] When looking at multiple chunks (511, 512, 513), content within the same paragraph is included in different chunks, or content corresponding to different headings is included together.

[0081] On the other hand, it can be seen that multiple chunks (521, 522, 523) are divided based on the document structure, and each chunk is divided to consist of header information and corresponding content information.

[0082] Product Information #2 (520) is the user manual for L100 (product), and the header information is "##1. Basic Information", "###1.1 Features", and "###1.2 Inverter Nameplate and Model Description", and the content information may be the content included between each header information. The metadata may include the title, product name, chunk creation date and time, chunk storage location, etc. of each header information.

[0083] For example, when the information providing device (100) receives a prompt inquiring about the features of L100, it can identify multiple chunks (521, 522, 523) having metadata corresponding to the prompt based on the prompt ("L100", "features"). The information providing device (100) can re-rank the multiple chunks (521, 522, 523) regarding L100 and generate result information based on the chunk (522) with the highest priority.

[0084] According to one embodiment of the present invention, the performance of a generative artificial intelligence model can be enhanced by using chunks that are divided by reflecting contextual information.

[0085] FIG. 6 is a diagram illustrating the operation of an information providing device according to an embodiment of the present invention.

[0086] As described above, the database (10) of FIG. 1 may be implemented as two or more databases as needed. The database may include a product database (11) and a vector database (12). At this time, each database may be implemented separately, but two or more databases may be combined as needed.

[0087] The information providing device (100) can collect product information and build a product database (11). As previously described, the product information can be divided into multiple chunks and stored, and each chunk can be composed of header information and content information. In addition, metadata provided corresponding to each chunk can be stored together.

[0088] The information providing device (100) can build a vector database (12) by vector embedding product information and metadata.

[0089] When the information providing device (100) receives a prompt (720), it can identify result information based on information stored in the product database (11) and the vector database (12) through a generative artificial intelligence model (710). At this time, embedding vectors or metadata can be utilized to identify result information corresponding to the prompt (720).

[0090] According to one embodiment of the present invention, an independent database for product information within a smart factory can be established to provide information that is closely related to target products or companies and is of high importance and practicality.

Claims

1. In an information providing device based on a generative artificial intelligence model, Receive a prompt regarding a product from at least one user terminal, and Identifying the first metadata corresponding to the above prompt among the metadata corresponding to product information divided into multiple chunks based on the document structure, and Based on the first metadata and a first plurality of chunks corresponding to the first metadata, result information corresponding to the prompt is generated, and An information providing device comprising a processor that transmits the above result information to the user terminal.

2. In Paragraph 1, The above processor is, An information providing device that generates result information based on similarity between the embedding vectors of the prompt and the first metadata and the embedding vectors of a plurality of chunks and metadata within a vector database using a generative artificial intelligence model.

3. In Paragraph 1, The above processor is, An information providing device that divides the above product information into multiple chunks based on a document structure.

4. In Paragraph 3, The above processor is, Identify header information within the above product information based on text formatting, and An information providing device that divides product information into multiple chunks by mapping header information and content information placed between header information.

5. In Paragraph 1, The above processor is, Re-rank the first plurality of chunks based on the similarity between the first plurality of chunks and the prompt, and An information providing device that generates the result information based on the first multiple reranked chunks.

6. Regarding the method of providing information based on a generative artificial intelligence model, A step of receiving a prompt regarding a product from at least one user terminal; A step of identifying a first metadata corresponding to the prompt among metadata corresponding to product information divided into multiple chunks based on a document structure; A step of generating result information corresponding to the prompt based on the first metadata and a first plurality of chunks corresponding to the first metadata; A method for providing information comprising the step of transmitting the above result information to the user terminal.

7. In Paragraph 6, The step of generating the above result information is, A method for providing information comprising the step of generating result information based on similarity between the embedding vector of the prompt and the first metadata and the embedding vector of a plurality of chunks and metadata within a vector database using a generative artificial intelligence model.

8. In Paragraph 6, A method for providing information that further includes the step of dividing the product information into multiple chunks based on the document structure.

9. In Paragraph 8, The step of dividing into a plurality of chunks is, A step of identifying header information within the product information based on text formatting; A method for providing information comprising the step of dividing product information into multiple chunks by mapping header information and content information placed between header information.

10. In Paragraph 6, The step of generating the above result information is, A step of re-ranking the first plurality of chunks based on the similarity between the first plurality of chunks and the prompt; A method for providing information comprising the step of generating result information based on a first plurality of reranked chunks.

Images