Information processing device
By chunking Japanese documents into paragraph units and associating them for numerical vector conversion, the information processing device improves search accuracy in chatbots by retrieving coherent chunks from the knowledge base, enhancing large language model responses.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2024-12-02
- Publication Date
- 2026-06-12
AI Technical Summary
The search accuracy of a knowledge base using Japanese-language documents in chatbots decreases due to the extraction of only partial chunks from linked paragraphs, affecting the accuracy of large language models' responses.
An information processing device that chunks Japanese documents into paragraph units, associates these chunks based on relationships, and converts them into numerical vectors for improved search accuracy by extracting relevant and associated chunks from the knowledge base.
Enhances the search accuracy of the knowledge base by retrieving multiple linked paragraphs as coherent units, thereby improving the accuracy of large language model responses.
Smart Images
Figure 2026095994000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information processing apparatuses.
Background Art
[0002] As this type of apparatus, for example, an apparatus has been proposed that generates query data based on a document for a language model and uses a pair of the document and the query data for learning a search model for a chatbot (see Patent Document 1).
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] As a chatbot, a chatbot using a mechanism (Retrieval-Augmented Generation: RAG) that gives an independent information source to a large language model by combining a large language model (Large Language Models: LLM) and search of a specific information source (hereinafter, appropriately referred to as a “knowledge base”) has been proposed. Here, the knowledge base may include a plurality of text data (for example, documents). The knowledge base may include Japanese-language documents as text data. For example, when a knowledge base including Japanese-language documents is used in the chatbot, the search accuracy of the knowledge base may decrease. Note that a large language model is a language model constructed using a very large dataset and deep learning technology.
[0005] The present invention has been made in view of the above problems, and an object thereof is to provide an information processing apparatus capable of improving the search accuracy of a knowledge base. [Means for solving the problem]
[0006] An information processing device according to one aspect of the present invention includes a search means for searching for a chunk related to a question sentence entered by a user from a database which is generated by chunking a Japanese document into paragraph units and which includes a plurality of chunks associated based on the relationships between paragraphs, wherein the search means obtains a search result which includes one chunk related to the question sentence and other chunks associated with the first chunk. [Brief explanation of the drawing]
[0007] [Figure 1] This is a diagram showing the configuration of an information processing system according to an embodiment. [Figure 2] This is a block diagram showing an example of a computer configuration. [Modes for carrying out the invention]
[0008] Embodiments relating to the information processing device will be described with reference to Figures 1 and 2. In Figure 1, the information processing system 1 comprises an information processing device 10, a server 20, a knowledge base 30, and an information processing device 40. The information processing device 10, server 20, knowledge base 30, and information processing device 40 are configured to communicate with each other via a network NW. Server 20 is a server for operating a large-scale language model (LLM). For this reason, server 20 may be referred to as an LLM server. Server 20 may be a cloud server.
[0009] The information processing devices 10 and 40 may be implemented by a computer COM as shown in Figure 2. In Figure 2, the computer COM comprises an arithmetic unit 110, a storage device 120, a communication device 130, an input device 140, and an output device 150. The arithmetic unit 110, storage device 120, communication device 130, input device 140, and output device 150 are connected via a data bus 160. The computer COM may be a personal computer, a tablet terminal, or a smartphone.
[0010] The arithmetic unit 110 may have a processor. The arithmetic unit 110 may have a single processor or multiple processors. In other words, the arithmetic unit 110 may have one or more processors. Furthermore, the processor may be a multi-core processor. If the arithmetic unit 110 has a single processor that is a multi-core processor, then logically, the arithmetic unit 110 can be said to have multiple processors.
[0011] The processor may be at least one of the following: CPU (Central Processing Unit), GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and TPU (Tensor Processing Unit).
[0012] The storage device 120 may be at least one of the following: RAM (Random Access Memory), ROM (Read Only Memory), hard disk drive, magneto-optical disk drive, SSD (Solid State Drive), and optical disk array. In other words, the storage device 12 may be implemented by a single device or by multiple devices.
[0013] The communication device 130 may be capable of communicating with devices outside of the computer COM. Furthermore, the communication device 130 may use either wired or wireless communication.
[0014] The input device 140 is a device capable of receiving information input to the computer COM from an external source. The input device 140 may include an operating device that can be operated by the user of the computer COM (e.g., a keyboard, mouse, touch panel, etc.). The input device 140 may include a recording medium reader capable of reading information recorded on a recording medium that can be attached to and detached from the computer COM, such as a USB (Universal Serial Bus) memory. Furthermore, when information is input to the computer COM via the communication device 130 (in other words, when the computer COM acquires information via the communication device 130), the communication device 130 may function as an input device.
[0015] Output device 150 is a device capable of outputting information to the outside of computer COM. Output device 150 may have a display device capable of outputting visual information such as characters and images as the above information. Output device 150 may also have a speaker capable of outputting auditory information such as sound as the above information. Output device 15 may have a vibration motor capable of outputting tactile information such as vibration as the above information. Output device 150 may also have a printer. Output device 150 may be capable of outputting information to a recording medium that can be attached to and detached from computer COM, such as a USB memory stick. When computer COM outputs information via communication device 130, communication device 130 may function as an output device.
[0016] The storage device 120 is capable of storing desired data. The storage device 120 may store the computer program CP that the arithmetic unit 110 will execute. The storage device 120 may temporarily store data that the arithmetic unit 110 will use temporarily when the arithmetic unit 110 is executing the computer program CP.
[0017] Furthermore, the computer program CP may be recorded on a non-temporary recording medium that is readable by a computer. In this case, the computer program CP may be stored in the storage device 120 by reading the recording medium using a recording medium reading device (not shown) provided by the computer COM. Furthermore, at least one of the following may be used as the recording medium: an optical disc, a magnetic medium, a magneto-optical disc, a semiconductor memory, and any other medium capable of storing a program. Furthermore, the computer program CP may be obtained from an external device (not shown) of the computer COM via a communication device 130. In other words, the computer program CP may be downloaded from an external device to the storage device 120 of the computer COM.
[0018] The arithmetic unit 110 (for example, a processor) may execute the processing that the computer COM should perform together with the memory device 120 in which the computer program CP is stored (in other words, together with the memory device 120 and the computer program CP stored in the memory device 120). For example, by the arithmetic unit 110 executing the computer program CP, a logical functional block for executing the processing that the computer COM should perform may be realized within the arithmetic unit 110 (for example, within the processor).
[0019] (Knowledge Base) Knowledge base 30 may contain multiple text data entries. The text data registered in knowledge base 30 may be vectorized text data. In other words, knowledge base 30 may be a vector database / vector store. In this embodiment, in particular, knowledge base 30 contains text data related to the Japanese language.
[0020] The information processing apparatus 40 shown in FIG. 1 may construct a knowledge base 30. The information processing apparatus 40 includes a splitting unit 41, an associating unit 42, and a conversion unit 43 in order to construct the knowledge base 30. The splitting unit 41, the associating unit 42, and the conversion unit 43 may be realized as the above-described logical functional blocks. Note that at least one of the splitting unit 41, the associating unit 42, and the conversion unit 43 may be realized as a physical processing circuit. Alternatively, at least one of the splitting unit 41, the associating unit 42, and the conversion unit 43 may be realized in a form in which logical functional blocks and physical processing circuits are mixed.
[0021] The text data registered in the knowledge base 30 may be, for example, fragmentary data obtained by splitting the text included in a document. The fragmentary data may be referred to as a “chunk”. The splitting unit 41 of the information processing apparatus 40 splits (i.e., chunks) a Japanese document represented in Japanese and generates a plurality of Japanese chunks. In the present embodiment, it is assumed that the splitting unit 41 splits the Japanese document in units of paragraphs.
[0022] The associating unit 42 of the information processing apparatus 40 associates the plurality of Japanese chunks based on the relationship between paragraphs. For example, examples of the “relationship between paragraphs” include introduction, development, change, and conclusion, introduction, main text, and conclusion, and the like. Information regarding association (for example, metadata) may be attached to each Japanese chunk. Note that not all of the plurality of Japanese chunks need to be associated. Note that the relationship between paragraphs may be estimated by a method such as dependency structure analysis.
[0023] The conversion unit 43 of the information processing apparatus 40 converts each of the plurality of Japanese chunks into a numerical vector. That is, the conversion unit 43 generates an embedding based on each of the plurality of Japanese chunks.
[0024] The information processing apparatus 40 registers the Japanese chunks converted into numerical vectors in the knowledge base 30. As a result, a plurality of Japanese chunks associated based on the relationship between paragraphs are registered in the knowledge base 30.
[0025] (Chatbot) Server 20 and Knowledge Base 30 may provide a chatbot service using RAG. For example, user U may use the chatbot service via information processing device 10. In this case, user U may operate information processing device 10 to launch an application for using the chatbot service. User U may operate information processing device 10 to enter a question into the input field of the chat application. Here, "question" is not limited to interrogative sentences. For example, "question" may be a sentence that includes expressions such as requests, instructions, or commands, such as "Tell me about ****" or "Answer me about ****". Therefore, "question" is a concept that includes not only sentences in the form of interrogative sentences, but also sentences that include expressions such as requests, instructions, or commands. In other words, "question" may mean a sentence that seeks an answer from the other party.
[0026] The search unit 11 of the information processing device 10 searches the knowledge base 30 based on the input question sentence. Specifically, the search unit 11 may convert the question sentence into a numerical vector. The search unit 11 may search the knowledge base 30 by comparing the question sentence converted into a numerical vector with numerical vectors (i.e., chunks) registered in the knowledge base 30. In this case, the search unit 11 may extract one Japanese chunk related to the question sentence from the knowledge base 30. If one Japanese chunk is extracted, the search unit 11 may extract other Japanese chunks associated with that Japanese chunk from the knowledge base 30. Then, the search unit 11 may obtain search results that include the one Japanese chunk and the other Japanese chunks.
[0027] The input unit 12 of the information processing device 10 may send first information to the server 20, which includes the input question and chunks as search results from the knowledge base 30 (for example, one Japanese chunk and another Japanese chunk). The server 20 may input the question and chunks included in the first information as prompts to the large-scale language model. The server 20 may obtain the answer to the question output from the large-scale language model. The server 20 may send second information indicating the answer to the information processing device 10. The information processing device 10, having received the second information, may display the answer indicated by the second information on a screen related to the chat application.
[0028] (Technical effects) From the perspective of ease of data handling, data is often registered in the knowledge base 30 as chunks (i.e., fragmented data). However, in Japanese documents, two or more paragraphs sometimes work together to express a single coherent meaning. If, for example, a search of the knowledge base 30 extracts only chunks corresponding to a part of two or more linked paragraphs, the accuracy of the large-scale language model's answer to the user U's question may decrease. In contrast, the information processing device 10 according to this embodiment may, when it extracts one Japanese chunk related to the question from the knowledge base 30, extract other Japanese chunks associated with that first Japanese chunk from the knowledge base 30. With this configuration, multiple Japanese chunks corresponding to two or more linked paragraphs can be extracted. In other words, the information processing device 10 according to this embodiment can improve the search accuracy of the knowledge base 30. As a result, the accuracy of the large-scale language model's answer can be improved.
[0029] Various aspects of the invention derived from the embodiments described above are described below.
[0030] An information processing device according to one aspect of the invention includes a database which is generated by chunking a Japanese document into paragraph units and which includes a plurality of chunks associated based on the relationships between paragraphs, and a search means which searches for a chunk related to a question sentence entered by a user, wherein the search means obtains a search result which includes one chunk related to the question sentence and other chunks associated with the one chunk. In the above embodiment, "knowledge base 30" corresponds to an example of a "database", and "search unit 11" corresponds to an example of a "search means".
[0031] The information processing device in the above embodiment may include input means for inputting the question text and the first chunk and the other chunk included in the search results into a large-scale language model. In the above embodiment, the "input unit 12" corresponds to an example of the "input means".
[0032] The present invention is not limited to the embodiments described above, and can be modified as appropriate without contradicting the gist or idea of the invention as can be read from the claims and specification as a whole. Information processing devices that involve such modifications are also included within the technical scope of the present invention. [Explanation of Symbols]
[0033] 1…Information processing system, 10, 40…Information processing device, 20…Server, 30…Knowledge base, 11…Search unit, 12…Input unit
Claims
1. The system generates Japanese documents by chunking them paragraph by paragraph, and includes a search mechanism that searches for chunks related to a user-entered question from a database containing multiple chunks linked based on the relationships between paragraphs. The search means obtains search results that include one chunk related to the question and other chunks associated with the first chunk. Information processing device.
2. The system includes input means for inputting the aforementioned question text and the one chunk and the other chunk included in the search results into a large-scale language model. The information processing apparatus according to claim 1.