Training method, device, electronic device and storage medium for knowledge discovery system

Abstract

The present application relates to the field of natural language processing technology, and provides a training method, device, electronic device and storage medium for a knowledge discovery system. The method includes: obtaining articles and user knowledge needs in a training set, wherein the articles and knowledge needs correspond to the same knowledge field; using a large language model to process the articles and generate summary information, wherein the summary information includes content summaries of the same knowledge field, association information between articles in the same knowledge field, and relevant information corresponding to each article; using a pre-built multi-agent to perform knowledge reasoning based on knowledge needs, content summaries, association information and relevant information to obtain knowledge discovery results; based on the knowledge discovery results and true labels, training the knowledge discovery system to obtain a trained knowledge discovery system. The present application solves the problem of high cost of knowledge discovery in the prior art.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to a training method, device, electronic device and storage medium for a knowledge discovery system. Background Art

[0002] At present, in the field of knowledge discovery, especially in scientific research and technological innovation, we are faced with the challenge of integrating scattered academic papers, data reports and expert opinions to generate novel scientific hypotheses and theories. The integration of interdisciplinary knowledge is one of the important ways of scientific and technological innovation. Knowledge graphs can be used to organize interdisciplinary knowledge and reveal potential knowledge associations. However, existing technologies often rely on in-depth analysis and long-term accumulated experience of experts in the field, which is not only time-consuming and labor-intensive, but also difficult to cover interdisciplinary knowledge discovery, making the cost of knowledge discovery of existing technologies high. Summary of the invention

[0003] In view of this, the present application provides a training method, device, electronic device and storage medium for a knowledge discovery system to solve the problem of high cost of knowledge discovery in the prior art.

[0004] In a first aspect of the present application, a method for training a knowledge discovery system is provided, comprising: obtaining articles and user knowledge needs in a training set, wherein the articles and the knowledge needs correspond to the same knowledge field; processing the articles using a large language model to generate summary information, wherein the summary information includes a content summary of the same knowledge field, association information between articles in the same knowledge field, and related information corresponding to each article; performing knowledge reasoning using a pre-built multi-agent based on the knowledge needs, content summaries, association information, and related information to obtain knowledge discovery results; and training the knowledge discovery system based on the knowledge discovery results and true labels to obtain a trained knowledge discovery system.

[0005] According to a second aspect of the present application, a training device for a knowledge discovery system is provided, including: an acquisition module, configured to acquire articles and user knowledge needs in a training set, wherein the articles and the knowledge needs correspond to the same knowledge field; a summary module, configured to process the articles using a large language model to generate summary information, wherein the summary information includes content summaries of the same knowledge field, association information between articles in the same knowledge field, and relevant information corresponding to each article; an inference module, configured to perform knowledge inference using a pre-built multi-agent according to the knowledge needs, content summaries, association information, and relevant information to obtain knowledge discovery results; and a training module, configured to train the knowledge discovery system based on the knowledge discovery results and true labels to obtain a trained knowledge discovery system.

[0006] According to a third aspect of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0007] According to a fourth aspect of the present application, a storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0008] At least one of the above technical solutions adopted in this application can achieve the following beneficial effects:

[0009] By obtaining articles and user needs in the training set, multiple articles in the training set are first processed through a large language model to obtain summary information, which includes content summaries of multiple articles in the same knowledge field, association information between articles in the same knowledge field, and related information expanded by each article. Then, based on the user's knowledge needs, the pre-built multi-agent is used to perform knowledge reasoning on the summary information to obtain knowledge discovery results, which means that new theories or new technical discoveries inferred in the corresponding knowledge field can be obtained based on the summary information. These new theories or new technical discoveries are all related to the content of the articles in the corresponding field. Then, the obtained knowledge discovery results can be compared with the corresponding true labels in the training set, and the knowledge discovery system is trained based on loss minimization to obtain a trained knowledge discovery system. In this way, the trained knowledge discovery system can be used to perform knowledge discovery on the knowledge needs input by the user and multiple articles, and obtain interdisciplinary fusion discoveries in the corresponding knowledge field, that is, technical information that promotes scientific progress and innovation. In this way, training data can be used to build a knowledge discovery system and then to the final application, forming a complete closed loop of knowledge discovery and application. The trained knowledge discovery system can be used to explore knowledge and integrate across disciplines to obtain hypotheses from new perspectives. This not only greatly reduces labor costs, but also can efficiently and accurately expand multi-domain knowledge and solve current disciplinary problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 It is a flowchart of a training method for a knowledge discovery system provided in an embodiment of the present application;

[0012] Figure 2It is a flowchart of another training method of a knowledge discovery system provided in an embodiment of the present application;

[0013] Figure 3 It is a structural schematic diagram of a training device for a knowledge discovery system provided in an embodiment of the present application;

[0014] Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0016] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0017] In addition, it should be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "includes..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0018] A training method and device for a knowledge discovery system according to an embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0019] Figure 1 It is a flowchart of a training method for a knowledge discovery system provided in an embodiment of the present application.

[0020] like Figure 1 As shown, the training method of the knowledge discovery system includes:

[0021] S101, obtaining articles in a training set and knowledge requirements of users, where the articles and knowledge requirements correspond to the same knowledge domain;

[0022] S102, using a large language model to process the article and generate summary information, the summary information includes a content summary of the same knowledge field, association information between articles in the same knowledge field, and relevant information corresponding to each article;

[0023] S103, based on the knowledge requirements, content summary, associated information and related information, using the pre-built multi-agent to perform knowledge reasoning and obtain knowledge discovery results;

[0024] S104, training the knowledge discovery system based on the knowledge discovery results and the true labels to obtain a trained knowledge discovery system.

[0025] Specifically, during the training process of the knowledge discovery system, articles from different knowledge fields and users' knowledge needs can be input into the knowledge discovery system at the same time, or the articles and knowledge needs of each knowledge field can be input into the knowledge discovery system separately for training. If different knowledge field contents are input at the same time, the knowledge discovery system can perform knowledge reasoning on the articles in different knowledge fields separately. Of course, if there are joint needs, multiple knowledge fields can also be cross-analyzed for knowledge reasoning.

[0026] It should be understood that the article can be a relevant academic paper or a related article containing academic knowledge. There is no limitation here. When the trained knowledge discovery system is applied for knowledge reasoning, the obtained article can be input by the user, that is, when the user has knowledge needs, the relevant article is input into the trained knowledge discovery system for processing. It can also be a pre-built relevant database for storing relevant texts in different knowledge fields. The system analyzes the knowledge needs input by the user, identifies the knowledge field, and then automatically obtains the text in the corresponding knowledge field in the database based on the identified knowledge field.

[0027] It should also be noted that knowledge needs can be clear knowledge needs, such as: analyzing the solution to a certain technical defect in a certain field, the system obtains clear relevant knowledge and solutions after analysis. It can also be a vague knowledge need: a certain disease, so that the system can obtain the latest medical research, treatment plans and patient management insights of the disease after processing. Of course, when using the trained knowledge discovery system for processing, the relevant information of the corresponding knowledge field can be directly input, and the system will analyze the key points based on the input relevant information to obtain relevant insights and solutions. Of course, this processing method has higher requirements for the system, so the system needs to be trained in many aspects, which will not be specifically expanded here.

[0028] Furthermore, when the article is processed using a large language model, the content summary obtained can be understood as a comprehensive summary of a knowledge field, and the generated related information can be understood as rough knowledge expansion information obtained based on each article, which may include concepts related to the article or related technical points.

[0029] Among them, the real labels can be manually pre-labeled, that is, the knowledge discovery results obtained through analysis by domain experts. The real labels can be saved in the training set and used to optimize system parameters, adjust reasoning strategies, etc., which can improve the efficiency and accuracy of the knowledge recommendation system.

[0030] In some embodiments, a large language model is used to process an article to generate summary information, including: extracting the text content of the article, processing the text content based on a first preset rule to obtain available text information of the article; segmenting the available text information into paragraphs based on a second preset rule to obtain segmentation results; summarizing each segmentation result separately to obtain a segmentation summary, summarizing each article based on the segmentation summary to obtain a summary of each article; and generating a content summary, associated information, and related information based on the summary.

[0031] Specifically, after the training set is input into the knowledge discovery system, the input article can be processed by a large language model to obtain summary information. In this process, the content of the article can be extracted first to obtain usable text information, and the usable text information can be segmented to obtain a complete paragraph segmentation result. The segmentation results are then summarized using the large language model to obtain a segment summary. Each article is then summarized based on the segment summary to obtain a summary of each article. In the corresponding field, based on the summary of each article, a content summary of all articles in the same knowledge field, associated information of each article knowledge, and related information corresponding to each article are generated.

[0032] It should be understood that multiple articles in different knowledge fields are only used as training sets to train the knowledge discovery system. In the actual application of the knowledge discovery system, only articles in one knowledge field can be obtained, that is, the knowledge field obtained based on the analysis of knowledge needs.

[0033] Among them, the first preset rule and the second preset rule can be pre-configured in the large language model. The first preset rule aims to clean the extracted text content to obtain usable text information. The large language model can perform semantic understanding and analysis on the text content, delete paragraphs or sentences that do not belong to the knowledge field, and of course, the text content can be cleaned or filtered based on other rules. The second preset rule aims to segment the available text information to obtain complete paragraphs or complete sentences to facilitate summarization and merging to generate summary information. The second preset rule can be set based on the paragraph length in the article or based on the semantic understanding of the large language model. There is no restriction on the specific situation, and the segmentation result can meet the semantic integrity.

[0034] According to the method of the above embodiment, the available text information can be extracted from the article through the first preset rule, and then the available text information can be divided into paragraphs through the second preset rule, so that the system can process the article in detail and obtain accurate processing results. After that, a summary of each article can be obtained based on the segmented summary of each article, and the article information in the current knowledge field can be comprehensively summarized based on the summary of each article to obtain summary information. In this way, the articles can be comprehensively analyzed to generate the association information between each article and the relevant information of each article. Knowledge reasoning combined with the content summary of the corresponding knowledge field can quickly obtain accurate knowledge discovery results.

[0035] In some embodiments, the text content is processed based on a first preset rule to obtain usable text information of the article, including: traversing the text content, filtering and retaining the text content with a font size within a preset threshold range to obtain initial text information; based on a preset text standard template, the initial text information is processed to obtain usable text information, and the text standard template includes removing redundant blank lines, processing punctuation, and processing segmentation errors.

[0036] Specifically, before using the first preset rule to process the text content, you can first use a text extraction tool to extract the text content, such as using Python's third-party library Fitz to extract PDF information to obtain text content. Of course, the input form of the article is not limited to PDF, and the tool is not limited to Python's third-party library Fitz. It only needs to be able to accurately extract the text content.

[0037] Furthermore, a first preset rule can be set based on the font size and text specification, that is, the text content can be traversed to obtain the font size in the text content, and the text content can be filtered according to the font size, and the text content with a font size within the preset threshold range is retained as the initial text information, and the text content with a font size outside the threshold range is regarded as not belonging to the main text content, and it is deleted, and the text content with a font size within the threshold range is retained to obtain the initial text information. Subsequently, the initial text information is processed based on the preset text specification to obtain usable text information, which can be understood as further filtering useless information. In this process, redundant lines can be removed, redundant punctuation can be removed, incorrect punctuation can be deleted or corrected, and segmentation errors can be processed, etc., to clean up useless information and obtain accurate usable text information for subsequent knowledge reasoning.

[0038] It should be noted that if the first preset rule is used to process the text content to obtain usable text information, the input form of the article can be unified, such as PDF, or other convenient text forms, and the font sizes of different input articles can be unified. For example, the main text of all articles is set to size 4, the first-level title is set to size 2, the second-level title is set to size 3, and the footnotes and documents are set to size 5. In this way, the preset threshold range can be set to font size 2-4. Of course, there is no restriction here.

[0039] It should also be noted that the preset threshold range and text standard template can be changed according to actual conditions, and can be set manually or automatically adjusted during system training.

[0040] According to the method of the above embodiment, the text content can be processed by the first preset rule to obtain usable text information to help clean and filter the text content and delete the messy and useless parts. The initial text information can be further processed using the text standard template to make the text content more standardized and easy to understand. In this way, the text content can be preliminarily screened and standardized to facilitate further analysis and processing, reduce the processing pressure of the knowledge discovery system, promote the training of the knowledge discovery system, and facilitate the trained knowledge discovery system to process the input and reason about the knowledge to obtain accurate knowledge discovery results.

[0041] In some embodiments, the available text information is segmented into paragraphs based on a second preset rule to obtain segmentation results, including: if the paragraph length corresponding to the available text information does not exceed the preset length, the available text information is segmented based on the paragraph length to obtain segmentation results; if the paragraph length exceeds the preset length, the available text information is segmented based on the entire sentence length to obtain segmentation results.

[0042] Specifically, after obtaining the available text information in the article, the paragraph segmentation can be performed according to the original paragraph of the article. However, if the original paragraph length of the article exceeds the preset length, it can be segmented based on the integrity of the sentence. The period can be used as the segmentation point. If the period segmentation length still exceeds the preset length, the comma can be used as the segmentation point. If the comma segmentation length still exceeds the preset length, it can be violently segmented. At this time, the sentence can be segmented based on semantic analysis, and the integrity of the sentence expression can be maintained as much as possible to obtain the segmentation result. This process can be continuously trained and improved during the training process of the system. Of course, a more appropriate segmentation method can also be selected based on the content form of the article. For example, if the article contains formulas, the integrity of the formulas can be guaranteed. The formula segmentation can be set to be unaffected by the preset length, or the formula can be compressed to achieve the best effect. The preset length is not limited here, and can be set based on the writing habits of papers in the knowledge field. For example, in the field of law, if UTF-8 encoding is used, the preset length can be set to 1000 bytes, which is not limited here.

[0043] According to the method of the above embodiment, the available text information in the article can be segmented, and the integrity and coherence of the paragraphs or sentences are guaranteed during the segmentation process, which can improve the readability and comprehensibility of the text. It not only makes the segmented text easier to be further processed by other modules, but also improves the usability and applicability of the text content, thereby effectively promoting the training of the knowledge discovery system to obtain accurate research results.

[0044] In some embodiments, the multi-agent includes: an information extraction agent, an analysis tool agent, a hypothesis generation agent, and a commentary agent.

[0045] Specifically, multiple agents can be constructed in the knowledge discovery system in advance, and different roles in the multiple agents can be played by a large language model, among which the information extraction agent, analysis tool agent, hypothesis generation agent and comment agent can be played by the same large language model or by different large language models respectively.

[0046] In other embodiments, knowledge reasoning is performed using a pre-built multi-agent to obtain knowledge discovery results, including: using an information extraction agent to extract key information from content summaries, associated information, and related information; using an analysis tool agent to select retrieval tools and retrieval resources, formulate and execute retrieval strategies, and obtain useful information, wherein the retrieval tools and retrieval resources are selected based on knowledge needs and key information, and the retrieval strategies are formulated based on the retrieval tools, retrieval resources, and key information; using a hypothesis generation agent to construct hypotheses of new perspectives based on key information and useful information; and using a comment agent to comment on the hypothesis to obtain knowledge discovery results.

[0047] Specifically, the information extraction agent can be used as an analyst to extract key information from summary information, i.e. content summary, associated information, and related information, and then guide the work of the analysis tool agent, where the key information can be obtained based on the current knowledge domain, for example, it can include important concepts, events, etc.

[0048] The analysis tool agent can be regarded as an engineer, selecting appropriate retrieval tools and retrieval resources based on knowledge needs, and formulating and executing retrieval strategies. In this process, the needs can be understood first, that is, the knowledge needs input by the user and the background of the knowledge needs can be analyzed to obtain the retrieval fields, topics or keywords that need to be retrieved. In this way, the appropriate retrieval tools and retrieval resources can be further determined, that is, the appropriate search engines, databases, document resources, etc. can be selected, such as Internet search engines, academic databases, library catalogs, etc., and then retrieval strategies can be formulated, such as determining search keywords, using logical operators to combine keywords and restrictions, etc., and executing the retrieval strategy to obtain useful information. Useful information may include references, document abstracts, extracted key information, etc. It should be understood that this process can automatically adjust parameters through the training of the knowledge discovery system so that the system can obtain accurate and useful information. It should also be noted that the knowledge discovery system can connect to external databases to execute retrieval strategies.

[0049] The hypothesis-generating agent can be used as a scientist to summarize and infer key and useful information to generate new insights or understandings, that is, to obtain new hypotheses.

[0050] The review agent can be used as a critic to analyze whether the new hypothesis obtained by the hypothesis generation agent meets the application requirements, practicality, novelty, etc. If it does, the new hypothesis will be output as the knowledge discovery result. If it does not, the knowledge reasoning will be re-performed.

[0051] It should be noted that human-computer interaction can be realized during the reasoning process of multiple agents. That is, if the operator turns on human-computer interaction, the output of each agent can be visualized, and the operator can judge whether the output is correct. If the output is wrong, the operator can directly edit the text information and then instruct the multiple agents to re-reason the knowledge based on the operator's text information or instruct the next agent to perform the operation based on the operator's text information, so as to achieve the accuracy of the knowledge discovery results.

[0052] In this way, multi-agent knowledge reasoning can help extract useful knowledge from large amounts of information, and conduct in-depth analysis and understanding, thereby providing support and guidance for decision-making, innovation, etc.

[0053] In some embodiments, a comment agent is used to comment on hypotheses to obtain knowledge discovery results, including: performing multi-angle analysis on the hypothesis based on preset prompt words in the comment agent, and if the hypothesis meets the prompt words, outputting the hypothesis as a knowledge discovery result; if the hypothesis does not meet the prompt words, instructing multiple agents to re-perform knowledge reasoning through the comment agent until the hypothesis meets the prompt words.

[0054] Specifically, prompt words can be used to comment on hypotheses. For example, in the medical field, the prompt words can be set to feasibility and accuracy. In this way, when the commentary agent processes the knowledge needs in the medical knowledge field, it will analyze the obtained hypotheses based on feasibility and accuracy. This process can be based on the available information retrieved by the analysis tool agent, or it can re-retrieve relevant information for analysis.

[0055] It should be noted that this process requires accurate semantic analysis and processing capabilities of a large language model. Therefore, a large amount of multi-agent training can be performed in advance to obtain efficient and accurate multi-agents, and then to obtain a trained knowledge discovery system.

[0056] According to the method of the above embodiment, the large language model can analyze the hypothesis based on the prompt words to obtain accurate knowledge discovery results, which can help train the knowledge discovery system and also assist the trained knowledge discovery system to obtain knowledge discovery results that effectively promote scientific and technological progress and innovation.

[0057] Figure 2 This is a flow chart of another training method for a knowledge discovery system provided in an embodiment of the present application. Figure 2 Further explanation of this application.

[0058] After the knowledge discovery system is trained, a user may input a collection of papers. Of course, the collection of papers may be stored in a database in advance, and the knowledge requirements input by the user may be analyzed to retrieve the collection of papers in the corresponding knowledge field.

[0059] Furthermore, the text content of the collection of papers is extracted to obtain the text content, which is then processed by a large language model to obtain a summary of each paper. Summary information containing information of multiple papers in the corresponding knowledge field is then generated, and the summary information is input into the multi-agent system for processing, and interacts with the user to obtain the final knowledge discovery result. In this way, knowledge discovery results that satisfy users or critics can be obtained, and analysis and reasoning of papers can be achieved. Through writing between agents, information from different disciplines can be integrated, potential connections and patterns can be identified, and new scientific hypotheses or theories can be proposed.

[0060] As some examples: When an enterprise R&D team explores new technology applications, team members can input industry trends and technical specifications through the system, and then use the multi-agent system to analyze relevant scientific literature and market reports to obtain innovative technology application solutions and potential market opportunities.

[0061] When interdisciplinary research teams are looking for new research intersections, research team members can input their research interests and questions in their respective fields. The system will automatically analyze interdisciplinary literature and data to identify and recommend possible research intersections and cooperation opportunities.

[0062] When policymakers and social scientists conduct research on social issues, they can input descriptions and key data of relevant social issues, use multi-agent systems to analyze large amounts of social science literature and case studies, and obtain evidence-based analysis reports and policy recommendations.

[0063] In research and clinical applications in the field of healthcare, relevant information about a specific disease or treatment method can be input, and the system will automatically extract key knowledge from the latest medical research and clinical trial reports, and analyze and reason to obtain the latest insights into disease diagnosis, treatment plans, and patient management.

[0064] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described one by one here.

[0065] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0066] Figure 3 Schematic diagram of a training device for a knowledge discovery system provided in an embodiment of the present application. Figure 3 As shown, the training device of the knowledge discovery system includes:

[0067] An acquisition module 301 is configured to acquire articles in a training set and knowledge requirements of users, where the articles and knowledge requirements correspond to the same knowledge domain;

[0068] The summary module 302 is configured to process the article using the large language model to generate summary information, which includes content summary of the same knowledge field, association information between articles in the same knowledge field, and relevant information corresponding to each article;

[0069] The reasoning module 303 is configured to perform knowledge reasoning using a pre-built multi-agent according to knowledge requirements, content summaries, associated information, and related information to obtain knowledge discovery results;

[0070] The training module 304 is configured to train the knowledge discovery system based on the knowledge discovery results and the true labels to obtain a trained knowledge discovery system.

[0071] In some embodiments, the summary module 302 is specifically used to extract the text content of the article, process the text content based on the first preset rule to obtain the available text information of the article; segment the available text information into paragraphs based on the second preset rule to obtain segmentation results; summarize each segmentation result separately to obtain a segmentation summary, summarize each article based on the segmentation summary to obtain a summary of each article; based on the summary, generate a content summary, related information and related information.

[0072] In some embodiments, the summary module 302 is specifically used to traverse the text content, filter and retain the text content with a font size within a preset threshold range, and obtain initial text information; based on a preset text standard template, the initial text information is processed to obtain usable text information, and the text standard template includes removing redundant blank lines, processing punctuation, and processing segmentation errors.

[0073] In some embodiments, the summary module 302 is specifically used to segment the available text information based on the paragraph length to obtain a segmentation result if the paragraph length corresponding to the available text information does not exceed a preset length; if the paragraph length exceeds the preset length, the available text information is segmented based on the entire sentence length to obtain a segmentation result.

[0074] In some embodiments, the reasoning module 303 is specifically used to utilize an information extraction agent to extract key information from content summaries, associated information, and related information; utilize an analysis tool agent to select retrieval tools and retrieval resources, formulate and execute retrieval strategies, and obtain useful information. The retrieval tools and the retrieval resources are selected based on knowledge needs and key information, and the retrieval strategies are formulated based on the retrieval tools, retrieval resources, and key information; utilize a hypothesis generation agent to construct hypotheses of new perspectives based on key information and the useful information; utilize a comment agent to comment on the hypotheses and obtain knowledge discovery results.

[0075] In some embodiments, the reasoning module 303 is specifically used to perform multi-angle analysis on the hypothesis based on the preset prompt words in the comment agent. If the hypothesis meets the prompt words, the hypothesis is output as a knowledge discovery result; if the hypothesis does not meet the prompt words, the comment agent is used to instruct the multiple agents to re-perform knowledge reasoning until the hypothesis meets the prompt words.

[0076] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0077] Figure 4 Schematic diagram of an electronic device 4 provided in an embodiment of the present application. Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 401 executes the computer program 403, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0078] The electronic device 4 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 4 may include, but is not limited to, a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 The electronic device 4 is merely an example and does not limit the electronic device 4 , and may include more or less components than those shown in the figure, or different components.

[0079] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0080] The memory 402 may be an internal storage unit of the electronic device 4, for example, a hard disk or memory of the electronic device 4. The memory 402 may also be an external storage device of the electronic device 4, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 4. The memory 402 may also include both an internal storage unit and an external storage device of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0081] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0082] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0083] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A training method for a knowledge discovery system, characterized in that: include: Acquire articles in a training set and knowledge requirements of users, wherein the articles and the knowledge requirements correspond to the same knowledge field; Processing the articles using a large language model to generate summary information, the summary information including a content summary of the same knowledge field, association information between the articles in the same knowledge field, and relevant information corresponding to each of the articles; According to the knowledge demand, the content summary, the associated information and the related information, a pre-built multi-agent is used to perform knowledge reasoning to obtain a knowledge discovery result, wherein an information extraction agent is used to extract key information from the content summary, the associated information and the related information; an analysis tool agent is used to select retrieval tools and retrieval resources, formulate and execute retrieval strategies, and obtain useful information, wherein the retrieval tools and the retrieval resources are selected based on the knowledge demand and the key information, and the retrieval strategies are formulated based on the retrieval tools, the retrieval resources and the key information; a hypothesis generation agent is used to construct a hypothesis of a new perspective based on the key information and the useful information; the hypothesis of a new perspective is a new insight or new understanding generated by summarizing and inferring the key information and the useful information; a comment agent is used to perform a multi-angle analysis on the hypothesis to obtain the knowledge discovery result, wherein the multi-agent includes the information extraction agent, the analysis tool agent, the hypothesis generation agent and the comment agent; Based on the knowledge discovery results and the true labels, the knowledge discovery system is trained to obtain a trained knowledge discovery system.

2. The method according to claim 1, characterized in that The article is processed using a large language model to generate summary information, including: Extracting text content of the article, and processing the text content based on a first preset rule to obtain usable text information of the article; Performing paragraph segmentation on the available text information based on a second preset rule to obtain a segmentation result; Summarizing each of the segmentation results respectively to obtain a segment summary, and summarizing each article based on the segment summary to obtain a summary of each article; Based on the summary, the content summary, the associated information and the related information are generated.

3. The method according to claim 2, characterized in that The processing of the text content based on the first preset rule to obtain the usable text information of the article includes: Traversing the text content, filtering and retaining the text content whose font size is within a preset threshold range, to obtain initial text information; The initial text information is processed based on a preset text standard template to obtain the usable text information, wherein the text standard template includes removing redundant blank lines, processing punctuation marks, and processing segmentation errors.

4. The method according to claim 2, characterized in that: The segmenting of the available text information into paragraphs based on the second preset rule to obtain the segmentation result includes: If the paragraph length corresponding to the available text information does not exceed a preset length, segmenting the available text information based on the paragraph length to obtain the segmentation result; If the paragraph length exceeds the preset length, the available text information is segmented based on the length of the entire sentence to obtain the segmentation result.

5. The method according to claim 1, characterized in that: The method of using the review agent to analyze the hypothesis from multiple angles to obtain the knowledge discovery result includes: Performing a multi-angle analysis on the hypothesis based on the preset prompt words in the comment agent, and outputting the hypothesis as the knowledge discovery result if the hypothesis satisfies the prompt words; If the hypothesis does not satisfy the prompt word, the comment agent instructs the multi-agent to re-perform knowledge reasoning until the hypothesis satisfies the prompt word.

6. A training device for a knowledge discovery system, characterized in that: include: An acquisition module is configured to acquire articles in a training set and knowledge requirements of users, wherein the articles and the knowledge requirements correspond to the same knowledge field; A summary module is configured to process the articles using a large language model to generate summary information, wherein the summary information includes a content summary of the same knowledge field, association information between the articles in the same knowledge field, and relevant information corresponding to each of the articles; The reasoning module is configured to perform knowledge reasoning using a pre-built multi-agent according to the knowledge demand, the content summary, the associated information and the related information to obtain a knowledge discovery result, wherein an information extraction agent is used to extract key information from the content summary, the associated information and the related information; an analysis tool agent is used to select a retrieval tool and a retrieval resource, formulate and execute a retrieval strategy, and obtain useful information, wherein the retrieval tool and the retrieval resource are selected based on the knowledge demand and the key information, and the retrieval strategy is formulated based on the retrieval tool, the retrieval resource and the key information; a hypothesis generation agent is used to construct a hypothesis of a new perspective based on the key information and the useful information; the hypothesis of a new perspective is a new insight or new understanding generated by summarizing and inferring the key information and the useful information; a comment agent is used to perform a multi-angle analysis on the hypothesis to obtain the knowledge discovery result, wherein the multi-agent includes the information extraction agent, the analysis tool agent, the hypothesis generation agent and the comment agent; The training module is configured to train the knowledge discovery system based on the knowledge discovery results and the true labels to obtain a trained knowledge discovery system.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images