Learning support system using search extension generation

Abstract

We provide a learning support system that uses search extension generation. [Solution] A learning support system that uses search-enhanced generation to present image data that is difficult to convert into text to the user and enables additional information to be added to the question text, comprising means for providing test questions, means for acquiring page image data, means for recording page text in a database, means for recording feature vectors in a database, means for extracting related feature vectors, means for removing duplicate reference addresses, and means for transcribing related text, wherein multiple RAG databases are constructed for each related field as needed for learning support to increase the storage capacity for additional information, and efficiency is improved by narrowing the target of RAG searches, and based on the chunks obtained from the RAG search, the text or image of the entire original document page in the part where the related content is written is referenced to eliminate reference omissions, and background information and related information that may be written around the searched chunks are reflected in the context of the question text.

Description

Technical Field

[0001] The present invention relates to learning support using search expansion generation, particularly to the understanding of test questions using options.

Background Art

[0002] In education at school and the like, it is a common practice to conduct a test after a lecture and have students answer it to test their understanding of the lecture content. For students with low scores, additional learning to supplement the insufficient parts of understanding is required, and for students with high scores, the provision of additional teaching materials to further deepen their understanding is required. In this regard, the application of machine learning is expected. In recent years, the development of machine learning has been remarkable, and particularly the spread of large language models (LLMs) has been progressing. Hundreds of billions to trillions of neural network parameters are learned with a large amount of data exceeding terabytes, and have been corresponding to tasks such as translation, speech and image recognition, and text summarization. In addition, generative AI that generates images, music, and documents based on instructions (prompts) is also being put into practical use. As it has become known that the performance improves by scaling the model (scaling law), the scaling of the model has been rapidly progressing.

[0003] As the application range of large language models expands, knowledge data in various fields are required, and with the progress of social situations and technologies in each field, the incorporation of always up-to-date knowledge data is demanded. However, learning a large amount of data requires large-scale computing resources, enormous power, and high costs. Therefore, the reconstruction of large language models cannot be performed frequently. When asking a large language model a question, there is a phenomenon called hallucination in which it generates an answer not based on facts even when the knowledge necessary to answer the question is not recorded internally. This is the reason why the spread is restricted in fields such as medicine where errors can directly lead to accidents.

[0004] To effectively utilize large-scale language models based on the latest information in real-world applications, additional up-to-date information is necessary, and currently, two approaches are being taken. One is called additional learning (fine-tuning) or transfer learning, where additional information is used to train a portion of the output layer (fine-tuning) or only the final layer (transfer learning) of an existing large-scale language model, creating a specialized large-scale language model. As it becomes a domain-specific large-scale language model, its usefulness is high. However, even if it is not as complex as training a large-scale language model from scratch, it still requires considerable expense and technical skills for training.

[0005] Another approach is the RAG Retrieval-Augmented Generation (RAG) approach used in this invention. The large-scale language model itself is not modified. Additional information is stored in a separate database (RAG database), and the information necessary to solve the question is retrieved from the RAG database. The obtained information is added to the question, and the large-scale language model is asked to produce an answer. Even if there is a large amount of potentially additional data, the additional information necessary to answer a given question is limited, so only that limited additional information is retrieved from the RAG database, and the extracted additional information (context) is added to the question in the question input field of the large-scale language model. In this approach, no changes are made to the large-scale language model itself through learning, and only contextual information is added to the question input field, resulting in lower costs and technical hurdles. Furthermore, it offers excellent responsiveness, such as being able to immediately reflect highly urgent information, such as urgent drug side effect information.

[0006] Here, the RAG database is created by breaking down the additional document information into small fragments (chunks), and then creating a database of feature vectors (chunk vectors) for each chunk. When a question is asked, the question itself is also vectorized, and chunk vectors with a high similarity to the question vector are searched from the RAG database. The contents of the resulting chunk group are then added to the question. This makes it possible to handle specific fields and the latest information without requiring additional training of expensive large-scale language models, and by limiting the basis for the answer to the question to the RAG database, it is possible to prevent hallucinations. The following are prior art documents related to this application. [Prior art documents] [Non-patent literature]

[0007] [Non-Patent Document 1] https: / / ja.wikipedia.org / wiki / %E5%A4%A7%E8%A6%8F%E6%A8%A1%E8%A8%80%E8%AA%9E%E3%83%A2%E3%83%87%E3%83%AB [Non-Patent Document 2] https: / / www.idnet.co.jp / column / page_308.html [Overview of the project] [Problems that the invention aims to solve]

[0008] Understanding learning objectives requires gathering information scattered across numerous books and documents. Traditionally, instructors manually collected information for each learning objective and then edited it into a format easily understandable to students. However, the sheer volume of learning objectives makes it difficult to provide sufficient information for every objective. To address this problem, the use of large-scale language models employing search-enhanced generation is expected to be crucial. However, when actually building a RAG database for learning support and attempting to search for information related to learning items, there are many unresolved problems, including: (1) the RAG database itself becomes enormous, resulting in a relatively large amount of data unrelated to the question; (2) the content to be searched is not always appropriately contained within chunks, leading to missed searches of data related to the question; (3) background information and related information that may be present around the searched chunk is omitted; and (4) current large-scale language models are still insufficient for image recognition and searching. Furthermore, repeatedly performing RAG searches on similar questions is a waste of resources such as cost and time. Furthermore, many exams present multiple options and ask candidates to select the appropriate one. Explaining all the options together often involves mixing different pieces of information, which not only increases the length of the explanation but also makes focused analysis difficult.

[0009] The present invention was made to solve the aforementioned problems of the past, and its purpose is to provide a learning support system using search-enhanced generation, which involves constructing multiple RAG databases for each related field as needed for learning support to increase the capacity for recording additional information, improving efficiency by narrowing the target of RAG searches, eliminating missed references by referencing the text or image of the entire original document page containing related content based on the chunks obtained from RAG searches, and at the same time reflecting background information and related information that may be written around the searched chunks into the context of the question, presenting image data that is difficult to convert into text to the user and enabling additional information to be added to the question, enabling focused explanations by breaking down questions with multiple choices into individual choices and generating answers, and eliminating the need for costly and time-consuming re-regeneration of answers by creating a database of question and answer records that can be searched. [Means for solving the problem]

[0010] As a means to achieve the above objective, the learning support system using the search extension generation described in claim 1 includes, in a large-scale language model, a means for providing test questions that record test questions and provide them as question sentences, (1) Page image acquisition means for acquiring image images of individual pages of information sources to be referenced separately from the large-scale language model during inference, along with the reference address to the image image. (2) A page text database recording means that extracts text strings from the page image and records them together with the reference address. (3) The extracted text is divided into small chunks, a feature vector database recording means is provided which calculates a feature vector and records it together with the reference address, (4) The large-scale language model is provided with a means for extracting related feature vectors, which extracts a group of feature vectors that are highly related to the test question, along with the reference address, from the feature vectors of the test question given to the test question. (5) The system includes a means for removing duplicate reference addresses from the extracted related feature vectors, (6) The system is characterized by having related text transcription means that extracts the page text specified by the reference address from the page text database and then transcribes it into the question input field together with the question text, thereby obtaining an answer to the test question by performing the operations (1) to (6) above.

[0011] The learning support system using search extension generation described in claim 2 is characterized in that, in the learning support system using search extension generation described in claim 1, it is further characterized by comprising a page image database recording means that records each acquired page image along with a reference address to the page image.

[0012] The learning support system using search extension generation according to claim 3 is characterized in that, in the learning support system using search extension generation according to claim 1 or 2, it is provided with page image viewing means for displaying and making available for viewing the page image specified by the reference address.

[0013] The learning support system using search extension generation according to claim 4 is characterized in that, in the learning support system using search extension generation according to claim 2 or 3, the large-scale language model is provided with a comment transcription means that extracts a page image specified by the reference address from the image data recording means, inputs comments to charts that have not been sufficiently converted to text, and transcribes them into the page text.

[0014] The learning support system using search extension generation according to claim 5 is characterized in that, in the learning support system using search extension generation according to claim 1 or 2, it comprises a plurality of page image acquisition means, a page text database recording means for the page image database recording means, and a plurality of feature vector recording means for the page text database recording means, and a plurality of RAG database search means for searching and extracting related feature vectors for any of the feature vector recording means.

[0015] The learning support system using search-extended generation according to claim 6 is characterized in that, in the learning support system using search-extended generation according to claim 1 or 2, it comprises a question-answer recording means for recording the question sentence and the obtained answer sentence, and for a new question sentence, the question-answer recording means first searches for a question sentence similar to the new question sentence, and if a similar question sentence is found, it comprises an F&Q database in which the answer record for the similar question sentence is used as the answer to the question.

[0016] The learning support system using search extension generation according to claim 7 is characterized in that, in the learning support system using search extension generation according to claim 1 or 2, the test question providing means includes a question text-option pair decomposition means for decomposing a question into pairs of question text and individual options. [Effects of the Invention]

[0017] In the learning support system using search expansion generation according to claim 1, since it is provided with examination question providing means, examination questions are recorded and provided as question texts. Since it is provided with page image image acquisition means, image images of individual pages of information sources to be separately referred to from the large language model during inference are acquired together with reference addresses to the image images. Since it is provided with page text database recording means, text of a character string is extracted from a page image image and recorded together with a reference address. Since it is provided with feature vector database recording means, the extracted text is divided into small segments (chunks) to calculate feature vectors and recorded together with reference addresses. Since it is provided with related feature vector extraction means, a group of feature vectors highly related to a test question are extracted together with reference addresses for the feature vector of the test question given to the large language model. Since it is provided with duplicate reference address removal means, duplicate reference address groups among the extracted related feature vectors are removed. Since it is provided with related text transcription means, after extracting page text specified by a reference address from the page text database, it is transcribed into a question input field together with a question text.

[0018] In the learning support system using search expansion generation according to claim 2, since it is provided with page image image database recording means, the acquired individual page image images are recorded together with reference addresses to the page image images.

[0019] In the learning support system using search expansion generation according to claim 3, since it is provided with page image image browsing means, the page image image specified by a reference address is used for display and browsing.

[0020] In the learning support system using the search extension generation described in claim 4, since it includes a comment transfer means, after extracting the page image image specified by the reference address from the image data recording means, comments are input for charts with insufficient text conversion and transferred to the page text.

[0021] In the learning support system using the search extension generation described in claim 5, it includes a plurality of page image image acquisition means, page text database recording means for the page image image database recording means, and the feature vector recording means for the page text database recording means. Furthermore, it includes a plurality of RAG database search means for searching and extracting related feature vectors for any feature vector recording means.

[0022] In the learning support system using the search extension generation described in claim 6, it includes a question sentence and a question-answer recording means for recording the obtained answer sentence. Furthermore, for a new question sentence, first, the question-answer recording means searches for whether there is a question sentence similar to the new question sentence. If there is a similar question sentence, it has a F&Q database that uses the answer record for the similar question sentence as the answer to the question.

[0023] In the learning support system using the search extension generation described in claim 7, the test question providing means includes a question sentence and option pair decomposition means for decomposing a question into a pair of a question sentence and individual options.

Brief Description of the Drawings

[0024] [Figure 1] This is an example of the network configuration of the present invention. [Figure 2] This is an example of a user interface in a large language model (LLM). [Figure 3] This shows the relationship between a large language model and a RAG database. [Figure 4] This shows the overall configuration of the extended RAG database of the present invention. [Figure 5]This is a detailed diagram of the process that constitutes the expanded RAG database. [Figure 6] This shows how reference addresses are numbered. [Figure 7] This is the procedure for searching the RAG database. [Figure 8] This is an example of an exam question. [Figure 9] This is an example of breaking down an exam question into a combination of the question statement and the answer choices. [Figure 10] Multiple RAG databases will be constructed for each domain. [Figure 11] This is an explanatory diagram of the F&Q database. [Modes for carrying out the invention]

[0025] Figure 1 shows a typical system configuration of the present invention. Since large-scale language models consist of massive amounts of data, numerous CPUs (Central Processing Units), GPUs (Graphics Processing Units), and a high-speed network connecting them, they are built within large servers such as cloud data centers and provided via the Web. Educational institutions such as schools have numerous PC terminals connected via a LAN (Local Area Network) that is connected to the web. Within these institutions, there are also servers that operate school administration databases managing lecture and exam content, student grades, and other related information. In recent years, there has been an increasing trend of using mobile devices such as smartphones and tablets to access cloud services, including large-scale language models, and school administration database servers from both inside and outside educational institutions.

[0026] Servers, terminals, and mobile terminals all consist of memory for storing programs and data, recording media such as hard disks for persistently storing the programs and data as needed, a CPU for reading and processing the programs and data, a GPU for high-speed parallel processing as appropriate, and communication modules. As cloud services become more stable and affordable, there is a growing trend of migrating some or all of the school administration database server functions to the cloud. Conversely, there is also a growing movement to move some or all of large-scale language models to devices with increased processing power and memory capacity (edge ​​computing). Furthermore, the development of small language models (SLMs), which have a limited number of parameters, is progressing. It should be noted that even these small models are still quite large compared to those developed before the emergence of large language models, and all embodiments, including this form, are included in the present invention.

[0027] Figure 2 shows an example of a user interface in a Large-Scale Language Model (LLM). LLM is currently under rapid development, with numerous models being created, including ChatG PT (a registered trademark of OpenAI), Bard, LaMDA (a registered trademark of Google), and LLaMA (a registered trademark of Meta). While the user interface will naturally differ, the standard configuration, as shown in Figure 2, consists of a prompt input field for entering instructions and inquiries to the L LM, a field for displaying the response to that prompt, and a field for displaying the history of prompts and responses as a usage log.

[0028] Recently, in addition to using LLM as a standalone tool as described above, there has been an increasing trend for LLM itself to have an API (Application Programming Interface), allowing its functions to be used by external software. In this case, prompts, responses, and history are input and output to and from external software via an API, so the display format is controlled by the external software and is not limited to Figure 2.

[0029] In large-scale language models, a one-hot vector is used to represent a vocabulary word. This vector consists of zeros with the same number of dimensions as the number of vocabulary words used, and a single 1 is placed at the position corresponding to that vocabulary word. The vocabulary of a large volume of literature is replaced entirely with vectors of this format, and the relationships (Attention) between each vocabulary vector are determined using deep learning. Then, for each query (prompt), the vocabulary that is most likely to follow the query and the already generated partial response is generated and added one word at a time to create the response. If the information used to generate the response is contained in the aforementioned large volume of literature, it is expected that a rational, useful, and correct sentence will be generated. However, if the information used to generate the response is not included in the aforementioned large volume of literature, the system will mechanically select vocabulary with a high probability of success, resulting in the generation of false responses that are not based on evidence (hallucination). When these hallucinations occur in settings such as healthcare and education, they can endanger patients' lives or cause students to believe in false information, which is one of the reasons why the application of large-scale language models to core business operations has not progressed.

[0030] Training large-scale language models requires massive servers, including a large number of parallel processing units (GPUs), as well as significant electricity resources and costs, to process large amounts of documents and use deep learning to determine relationships between vocabulary words. New documents are created every day, but it's not realistic to reflect all of them in a large-scale language model without any time lag. Furthermore, while the vast amount of documents scanned are primarily publicly available documents published on the web and other platforms, they do not include sensitive information such as internal company documents or electronic medical records from hospitals. Therefore, it is said that the documents that can be collected represent only a small fraction of all documents existing on Earth.

[0031] To effectively utilize large-scale language models based on the latest information in real-world applications, additional up-to-date information is required, and currently, two approaches are being taken. One approach is called additional learning (fine-tuning) or transfer learning, which involves using additional information to train a portion of the output layer (fine-tuning) or only the final layer (transfer learning) of an existing large-scale language model, thereby creating a specialized large-scale language model. While these domain-specific, large-scale language models are highly useful, they still require a considerable amount of investment and technical skill, even if it's not as complex as training a large-scale language model from scratch. Furthermore, because sensitive information such as personal details and descriptions of medical conditions included in the additional information is used for training, there is a risk that it may be accessed outside the organization. To prevent this, it is necessary to build and operate a large-scale language model with additional training exclusively within the company or hospital.

[0032] Another approach is the Search-Augmented Generator (RAG) approach used in this invention. The model of the large-scale language model itself is not changed. Additional information is stored in a separate database (RAG database), the information necessary to solve the question is retrieved from the RAG database, the text information of the obtained chunks is added to the question, and the large-scale language model is asked to produce an answer. Even if there is a large amount of potentially additional data, the additional information required to answer a given question is limited. Therefore, only that limited additional information is searched for in the RAG database, and the extracted additional information is added to the question (context) in the question input field of the large-scale language model. This approach involves no modification of the large-scale language model itself through training; it simply adds contextual information to the question input field, resulting in lower costs and technical hurdles. Furthermore, it offers excellent responsiveness, allowing for immediate reflection of urgent information such as emergency drug side effect information.

[0033] Figure 3 shows the flow of large-scale language model processing using a typical Search-Augmented Generative (RAG) method. A broker program, such as a chat application, receives a question from a user (1) and searches the RAG database for information related to the question (2). The broker program questions the LLM (4) with the obtained search results (3) along with a question, and receives a response from the LLM (5). It then presents the received response to the user (6). However, although the length of a chunk's text string is variable in the settings, there are certain limitations, and it may not be able to include all the necessary information. Conversely, if the chunks are too long, the focus of the feature vectors may become blurred, potentially preventing them from being properly handled by the search function. Also, if keywords overlap the transitions between chunks, proper feature vectorization may not be possible, making searching difficult. Furthermore, scanned images such as pictures and illustrations may not be fully recognized and converted into text, and therefore may not be found in searches. Furthermore, building a RAG database that is robust enough for practical use would require a massive system based on a huge number of document chunks. The necessary set of documents varies greatly depending on the area of ​​interest. For example, the types of documents required in the medical field are vastly different from those required in history or literature. Building a RAG database from a collection of documents covering all fields would result in a redundant configuration, potentially placing an unnecessary burden on database construction and searching.

[0034] Figure 4 shows the overall configuration of the extended RAG database according to the present invention. Figure 5 is a detailed diagram of the processing flow shown in Figure 4. From potentially additional information sources (information resources) such as books, ebooks, and PDF versions of web documents, scanned images of books and image images of PDF versions of web documents are recorded in a database (page image database) with a reference address such as "information resource name + page" assigned to each individual page image (page image image). Next, text is extracted from each page image (page text), and the page text is recorded along with the aforementioned reference address (page text database). The text on each page is divided into small chunks, a feature vector is calculated for each chunk, and the aforementioned reference address is assigned to record it in the feature vector database. Furthermore, if necessary, such as when there are strict limitations on recording capacity, the aforementioned page text may be compressed and recorded using the summarization function of a large-scale language model.

[0035] Here, the reference address is formatted as a book title followed by page numbering, as is common in books, as shown in Figure 6. However, unlike books which have physical constraints, there is no need to be particular about the display format in documents on the web, so depending on the granularity of the information, it may be numbered at the paragraph level, or conversely, at the section or chapter level. Page text can be unstructured plain text, but for complex content, tagged notation such as XML, JSON, or Markdown is preferable to clearly indicate the document's structure. Different software programs recommend different notations for calculating feature vectors. While this invention uses JSON notation, any notation may be used.

[0036] Currently, we are unable to perfectly represent images, videos, illustrations, and other content included in the page images as text. Therefore, ultimately, human judgment and understanding are required. It is also useful for users to add explanatory text to the corresponding page text based on their own judgment regarding the aforementioned images, videos, illustrations, etc., as needed. Thus, while a database of page images after extracting page text is not essential, it can be useful, and you can refer to the relevant page images as needed.

[0037] Figure 7 shows the search procedure for the RAG database. When a question is entered into the prompt input field, the broker program calculates the feature vector of the question and extracts a list of approximate feature vectors from the vector database (RAG database). The degree of approximation between feature vectors is often expressed by the magnitude of the dot product between the vectors (cosine similarity), but the Manhattan distance, which is the sum of the absolute differences of each element of the two vectors, can also be used. The size of the approximation threshold used as the basis for extraction is variable depending on the situation. If the extracted list is too large, you can raise the criteria to narrow it down; if the list is too small, you can lower the criteria and increase the size of the list. In some cases, you can set a threshold, such as the top 10 items, and automatically adjust the criteria so that the number of items extracted matches the set value.

[0038] In traditional RAG, when a list of feature vectors corresponding to multiple chunks was extracted, the text of the chunk from which each feature vector originated was added to the question text. However, as mentioned earlier, there are limitations on the length of the chunk's text string, and sometimes it was not possible to include all the necessary information. Furthermore, when keywords were applied to the transitions between chunks, proper feature vectorization was not possible. Furthermore, scanned images such as pictures and illustrations sometimes did not have sufficient text conversion. To compensate for this drawback, the present invention employs the following procedure.

[0039] The system organizes the reference addresses assigned to each feature vector, removes duplicate reference addresses, and transcribes the entire page text indicated by the remaining reference addresses into the prompt input box along with the question, prompting the large-scale language model to respond based on this. This procedure resolves the aforementioned problem with fragmented chunks, specifically the issue of gaps where each chunk cannot contain all the necessary information, while also allowing the text between chunks to be added to the question. This feature is useful in educating students and others by not only providing solutions to specific problems but also presenting the background information behind those problems, thereby promoting a deeper understanding.

[0040] If images, videos, illustrations, etc., are not adequately represented in text, the system will search for the image of the page in question from the page image database and display it to the user. The user will then add a description of the image, video, or illustration to the question. If necessary, this description may also be added to the page text. This allows for the utilization of information that was not previously fully utilized, such as images, videos, and illustrations. As described above, by following the procedure shown in Figure 7, when a learning item is questioned, information is collected from a wide range of information resources, including numerous books and documents, provided to the LLM, and organized and summarized. This makes it easier to create explanatory texts, which previously relied on time-consuming manual work.

[0041] Figure 8 shows an example of a typical exam question. While it would be acceptable to simply pose the question as is and request answers and explanations, the presence of numerous independent items makes it likely that redundant explanations will be generated. Furthermore, since there are countless combinations of choices, the explanatory text for each question tends to be repetitive. To avoid this, as shown in Figure 9, by decomposing the question into pairs of questions and individual choices, and then querying each individual choice, a focused and concise explanatory text can be expected (question-choice pair decomposition method). With this structure, it is easy to search for options similar to the one in question, and it is easy to create review questions to check whether the student has a sufficient understanding of the option they got wrong.

[0042] Building a RAG database that is robust enough for practical use requires a large-scale system based on a massive collection of document chunks. For example, in the medical field alone, there are well over 50 major specialties such as internal medicine, surgery, and obstetrics and gynecology, and these specialties are becoming increasingly subdivided as medicine advances. Attempting to consolidate all of these into a single RAG database would place a tremendous burden on data storage and retrieval. The necessary set of documents varies greatly depending on the area of ​​interest. For example, even within the medical field, there is considerable overlap between the documents used in abdominal surgery and those used in psychiatry. Furthermore, the collection of documents necessary for medicine differs significantly from those in the fields of history and literature. Therefore, as shown in Figure 10, by constructing multiple RAG databases for each area of ​​interest, and having the broker program search one or more RAG databases related to the question as needed, it is possible to avoid the inefficiency of searching due to the large size of the RAG database. Of course, if the RAG database becomes too fragmented, it will require performing numerous searches, which will reduce the efficiency of the search. Therefore, it goes without saying that it is necessary to build the RAG database by aggregating documents from highly relevant fields.

[0043] In educational settings, when generating explanations for exam answers, a general RAG database-based explanation may be appropriate for students who answered a question incorrectly and whose performance is not considered good. However, this risks failing to satisfy the intellectual curiosity of high-achieving students. In this case, it would be useful to build a RAG database containing fundamental and more advanced content, and then generate and provide in-depth explanatory texts from it.

[0044] While it's possible to create the questions and answers each time, this would require considerable computing resources and expense. As shown in Figure 11, pairs of questions and their corresponding answers are stored in the RAG database as an FAQ database. When a question is entered, the system first searches the FAQ database for answers that are closely related to the question. Only if the content of the found answers is unsatisfactory, the system creates a new answer using the procedure of the present invention and registers the result in the FAQ database. In this case, breaking down the problem into options as shown in Figure 9, rather than using the format in Figure 8, is more likely to yield a more versatile and non-redundant explanation. When registering for this FAQ database, you can either register only the feature vectors of the question text in the RAG database and link to the answer text, or you can chunk the question and answer text pairs and register the feature vectors for each chunk.

[0045] Furthermore, since there is a risk of information leakage if the content entered in the prompt is used to train a large-scale language model, it is useful to explicitly declare in the prompt that training is prohibited, or to use a paid version of a large-scale language model that guarantees that the content will not be used for training.

[0046] Although embodiments have been described above, the specific configuration of the present invention is not limited to the embodiments described above, and design changes and the like that do not depart from the spirit of the invention are also included in the present invention. For example, although the question text is provided by the test question provisioning means, the test questions or learning items may also be entered directly into the prompt input field. While this invention mainly describes a chunk feature vector database, the invention also includes adding information extracted from a regular relational database using SQL statements, or reference information from the web, to the prompt as appropriate, as long as it is useful for generation.

Claims

1. In a large-scale language model, a means for providing test questions is provided that records test questions and provides them as question texts. (1) Page image acquisition means for acquiring image images of individual pages of information sources to be referenced separately from the large-scale language model during inference, along with the reference address to the image image. (2) A page text database recording means that extracts text strings from the page image and records them together with the reference address. (3) The extracted text is divided into small chunks, a feature vector database recording means is provided which calculates a feature vector and records it together with the reference address, (4) The learning support system using the search extension generation method is provided with a related feature vector extraction means that extracts a group of feature vectors that are highly related to the test question, along with the reference address, from the feature vector of the test question given to the test question, (5) The system includes a means for removing duplicate reference addresses from the extracted related feature vectors, (6) The system includes related text transfer means that extracts the page text specified by the reference address from the page text database and then transfers it to the question input field together with the question text, A learning support system using search-enhanced generation, characterized by obtaining an answer to the test question by performing the operations (1) to (6) described above.

2. The learning support system using search extension generation according to claim 1, further comprising a page image database recording means that records each acquired page image along with a reference address to the page image.

3. A learning support system using search extension generation according to claim 1 or 2, characterized by comprising a page image viewing means for displaying and providing access to a page image specified by the aforementioned reference address.

4. A learning support system using the search extension generation described above, characterized in that it includes a comment transfer means for extracting a page image specified by the reference address from the image data recording means, inputting comments for charts that have insufficient text conversion, and transferring them to the page text, as described in claim 2 or 3.

5. A learning support system using search extension generation according to claim 1 or 2, characterized in that it comprises a plurality of page image acquisition means, a page text database recording means for the page image database recording means, and a plurality of feature vector recording means for the page text database recording means, and a plurality of RAG database search means for searching and extracting relevant feature vectors for any of the feature vector recording means.

6. A learning support system using search-extended generation according to claim 1 or 2, comprising a question-answer recording means for recording the aforementioned question and the obtained answer, wherein, with respect to a new question, the question-answer recording means first searches for a question similar to the new question, and if a similar question is found, the answer record for that similar question is used as the answer to the question, comprising an F&Q database.

7. The learning support system using search extension generation according to claim 1 or 2 is characterized in that the test question providing means comprises a question text-option pair decomposition means for decomposing a question into pairs of question text and individual options.

Images