A High-Value Information Mining Method Based on Expert Thinking Chain Large Model Intelligent Agent

Abstract

This invention discloses a high-value information mining method based on an expert thought chain large-scale model intelligent agent. The method involves inputting the task name and overview, extracting entity and event information keywords using a large language model, and vectorizing these keywords with an expert thought chain template. The cosine distance similarity between the keyword and the template vectors is calculated to obtain the most similar expert thought chain template. A search engine is then invoked to search for entity and event information keywords extracted from the task name and overview. For each keyword, the large model and search engine are used to associate the keyword with related content, obtaining and recording the title, publishing organization, publication time, content, and URL of the searched web pages. The recorded title, publication structure, and content are extracted to obtain entity, relation, and event triples, which are then organized into a public safety domain knowledge system for storage. Based on the expert thought chain template, the constructed public safety domain knowledge system is intelligently analyzed to obtain visualized statistical charts and structured analysis conclusions. Based on the visualized statistical charts and structured analysis conclusions, an analysis report is generated, resulting in the final information mining analysis report. This invention can quickly identify, classify, and parse large and complex datasets, thereby accelerating the information extraction and knowledge discovery process.

Description

TECHNICAL FIELD

[0001] The present application relates to natural language processing, and relates to a high-value information mining method based on an expert thinking chain large model agent. BACKGROUND

[0002] With the rapid development of information technology, the information field is facing the challenges of massive data processing and high-value information mining. High-value information mining based on large language models is a hot research direction in the field of natural language processing. Large language model agents are artificial intelligence models trained based on large model data sets, which can understand and generate natural language by learning the statistical rules of language, and can capture the complexity and diversity of information in the public security field. Although large language model agents have achieved unprecedented results in many tasks, their working mechanism lacks transparency and explainability, which can lead to the illusion problem.

[0003] Expert thinking chain refers to a series of logical reasoning and information mining steps adopted by experts in solving problems, which can simulate the expert thinking process through structured knowledge representation and reasoning rules, thereby improving the accuracy and reliability of the agent in solving problems. The method of expert thinking chain can significantly improve the performance of large language models in common sense question answering, logical reasoning and other tasks.

[0004] Therefore, guiding the agent to face the public security field for logical reasoning and information mining in a way specific to the information field can improve the agent's autonomous analysis and decision-making capabilities and provide new solutions to complex analysis problems. In information mining tasks, there are currently many problems such as the lack of effectiveness of intelligent information mining methods based on expert experience, the inability to solidify and backtrack expert experience in information mining tasks, and the difficulty of effectively reusing high-value information mining behaviors, which have become a pain point in the field of information mining research. This is mainly because expert experience knowledge, analysis patterns, and cognitive thinking were not included in the basic design at the beginning of the design of existing information systems. By combining large language models with expert thinking chains for high-value information mining, the accuracy and effectiveness of large language models in information mining can be effectively improved. At the same time, based on the guidance of expert thinking chains in the public security field, the high-value information mining capabilities and excellent natural language processing capabilities of "AI Agent + large model" are utilized to improve the quality and efficiency of information intelligent processing. SUMMARY

[0005] The present application aims to propose a high-value information mining method based on an expert thinking chain large model agent in the field of public security to solve the problem of insufficient effectiveness of intelligent information mining and processing methods based on expert experience.

[0006] The technical solution for achieving the object of the application is: a high-value information mining method based on an expert thinking chain large model agent, comprising the following steps:

[0007] Step 1, input the name and summary of the task, extract the entities including persons, times, places, and event information keywords from the large language model, and perform text vectorization processing on the keywords and the expert thinking chain template, match the expert thinking chain template with the highest similarity by calculating the cosine distance similarity between the keywords and the expert thinking chain template vectors;

[0008] Step 2, call the search engine to search for the entities and event information keywords extracted from the task name and summary, respectively use the large model and search engine to associate keywords and search related content for each information keyword, get and record the title, publishing agency, publishing time, content, and website address of the searched web page, extract the title, publishing structure, and content to get entity, relationship, and event triples, organize them into a public security field knowledge system for storage;

[0009] Step 3, according to the expert thinking chain template, intelligently analyze the constructed public security field knowledge system to obtain visual statistical charts and structured analysis conclusions;

[0010] Step 4, according to the visual statistical charts and structured analysis conclusions, write and generate an analysis report to obtain the final information mining analysis report.

[0011] Further, in step 1, input the name and summary of the task, extract the entities including persons, times, places, and event information keywords from the large language model, and perform text vectorization processing on the keywords and the expert thinking chain template, match the expert thinking chain template with the highest similarity by calculating the cosine distance similarity between the keywords and the expert thinking chain template vectors, wherein:

[0012] The expert thinking chain template includes two types of human events and natural disasters, wherein:

[0013] The human event template is: (1) please analyze the cause of the event; (2) please analyze whether the event is a domestic event or an international event; (3) please analyze the impact of public awareness in the event; (4) please analyze the strategy and guiding ideology of public security events; (5) please analyze the hard support of cognitive confrontation in the event; (6) please analyze the historical background and era characteristics of the event;

[0014] The natural disaster template is: (1) please analyze the cause of the event; (2) please analyze the international relations and diplomatic factors of the event; (3) please analyze the influence of public awareness in the event; (4) please analyze the strategy and guiding ideology of public safety events, disaster emergency plans, government and rescue organization strategies and response measures; (5) please analyze the hard support of cognitive confrontation in the event; (6) please analyze the historical background and characteristics of the times in the event.

[0015] Further, step 3, according to the expert thinking chain template, the constructed public safety field knowledge system is intelligently analyzed, and visual statistical charts and structured analysis conclusions are obtained, wherein:

[0016] Intelligent analysis includes visual data and structured analysis. In the visual data part, column chart is made for data containing different categories or time points and numerical size, line chart is made for data containing time change, radar chart is made for data containing relative size of different variables, trajectory chart is made for data containing state or position change at different time points, time axis is made for data containing time sequence and event sequence, pie chart is made for data containing proportion of each part to the whole. Six kinds of visual statistical charts are made, including column chart, line chart, radar chart, trajectory chart, time axis and pie chart. In the structured analysis part, star burst method, start list method or quadrant processing method are called to analyze and obtain structured analysis method conclusions.

[0017] Further, it also includes recording operation logs, which include two parts: behavior action and behavior content. Behavior action includes search, analysis and writing. Behavior content includes search content, search results, statistical chart data and structured analysis conclusions obtained by analysis, and written analysis report.

[0018] Further, it also includes adding the behavior action of obtaining high value information to the expert thinking chain template.

[0019] A high-value information mining method based on an expert thinking chain large model agent is provided.

[0020] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the high-value information mining method based on the expert thinking chain large model agent is implemented, and the high-value information mining based on the expert thinking chain large model agent is realized.

[0021] A computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the high-value information mining method based on the expert thought chain large model agent, realizes the high-value information mining based on the expert thought chain large model agent.

[0022] Compared with the prior art, the present application has the following advantages: 1) using the expert thought chain to guide the large model agent to carry out high-value information mining in the field of public safety, providing an efficient technical means for the field of information processing, and significantly improving the work efficiency of information processing personnel. Through the algorithm driving and data processing capability of multiple agents, a large number of complex data sets can be quickly identified, classified and analyzed, thereby accelerating the process of information extraction and knowledge discovery; 2) through the implementation of interactive operation mechanism, and combining with the auxiliary report generation function of the agent, the experience of the experts can be effectively solidified and traced back. This method not only promotes the reuse of high-value information mining behavior, but also enhances the knowledge base of information mining through intelligent recording and feedback cycle, providing rich reference resources and decision support for future information mining. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a high-value information mining method based on an expert thought chain large model agent structure diagram;

[0024] Figure 2 is an expert thought chain template sample diagram;

[0025] Figure 3 is an expert thought chain solidification iteration schematic diagram. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0027] The present application is directed to the field of public safety, wherein public safety refers to accidents or emergency situations that can have a significant impact on the life, health, property or environment of the public. High-value information mining based on expert thinking chain large model agent involves a central control system and a large model agent (hereinafter referred to as agent), the central control system is mainly used for interaction with the user and task allocation to the corresponding agent according to the decomposition, the agent part includes search agent, extraction agent, analysis agent and writing agent, each type of agent includes four core components of tool, memory, planning and action, tool refers to algorithms and models for processing and analyzing information data, memory refers to the ability of the agent to store information, planning refers to the action planning scheme formulated by the agent according to the target and the current environment state, and action refers to the response of the agent to the information mining task. The search agent tool mainly includes search-related tools such as searchers, and its main function is to search and associate keywords based on the task name and keywords in the summary input by the user; the extraction agent tool mainly includes extraction-related tools such as extraction large models, and its main function is to extract structured information such as entities, relationships and events based on the task name and summary input by the user and the text information searched by the search agent; the analysis agent tool includes a code compiler, and its main function is to analyze and process the text information searched by the search agent and the structured information extracted by the extraction agent, mainly including generating statistical charts and analyzing text, containing numerical information such as comparison and change, then using the code compiler to generate statistical charts, calling corresponding visual analysis interfaces according to common structured analysis methods in the information mining field (such as star burst method, start-up checklist method, quadrant processing method, etc.), and generating corresponding analysis results according to the structured analysis method; the writing agent tool mainly includes a writing large model, and its main function is to use the language generation capability of the writing large model to generate search information, structured knowledge and statistical charts generated by the search agent, extraction agent and analysis agent, and conclusions generated by the analysis agent, and to intelligently generate and write analysis reports.

[0028] As Figure 1 shown, a high-value information mining method based on expert thinking chain large model agent, comprising the following steps:

[0029] Step 1, task matching with expert thinking chain:

[0030] Firstly, the user inputs the "name and outline of the task", which is composed of text. According to the input task name, the text content of the outline, the large language model is used to extract the entity and event information keywords such as characters, time and place, and the keywords and expert thinking chain templates are processed by text vectorization. The cosine distance similarity between the keyword vector A and the expert thinking chain template vector B is matched by calculating the cosine distance similarity θ between the two vectors. A i and B i represent the components of vectors A and B respectively. The calculation method of cosine similarity is as follows:

[0031]

[0032] The cosine similarity is sorted from high to low, and the expert thinking chain template with the highest cosine similarity is obtained as the matching result of the expert thinking chain template. The initial expert thinking chain template is shown in Figure 2 , which is constructed by artificial experience. The artificial constructed expert thinking chain template for public security field includes two categories of human events and natural disasters (such as earthquakes, floods, hurricanes, etc.). The human event template is: (1) please analyze the cause of the event, such as safety hazards, illegal operations, etc.; (2) please analyze whether the event is a domestic event or an international event; (3) please analyze the influence of public cognition in the event; (4) please analyze the strategy and guiding ideology of public security events, such as the goals, strategies and ideological basis of the event organizers; (5) please analyze the hard support of cognitive confrontation in the event, such as media tools, network platforms, and propaganda channels; (6) please analyze the historical background and characteristics of the times in the event, such as similar events in history and the background of the times. The natural disaster template is: (1) please analyze the cause of the event, such as natural factors, such as geological activity, weather change, environmental deterioration, etc.; (2) please analyze the international relations and diplomatic factors of the event, such as international rescue cooperation, cross-border resource mobilization, international organization assistance and support; (3) please analyze the influence of public cognition in the event, such as media reports, information dissemination on social media; (4) please analyze the strategy and guiding ideology of public security events, disaster emergency plan, government and rescue organization strategy and response measures; (5) please analyze the hard support of cognitive confrontation in the event, such as emergency information system, media propaganda, information release channel; (6) please analyze the historical background and characteristics of the times in the event, such as lessons learned from similar disasters in the past, trends of current climate change and its impact.

[0033] After obtaining the matched expert thinking chain template, the central control system will decompose the task and assign the search, extraction, analysis and writing tasks to the corresponding search, extraction, analysis and writing agents to complete the search, extraction, analysis and writing tasks respectively.

[0034] Step 2, construction of search agent and knowledge system:

[0035] The search agent calls the search engine to search for the information keywords extracted from the task name and summary according to the matched expert thinking chain content in step 1. The big model and the search engine are used to associate 10 keywords for each information keyword and search related content. The title, publishing agency, publishing time, content, website address, and other attribute information of the searched news, information, and announcement web pages are obtained and recorded. The recorded title, publishing structure, and content attribute text information are transmitted to the extraction agent for extraction to obtain entity, relationship, and event information. The extraction process relies on the extraction function of the existing large language model to convert unstructured text information into structured information such as triples. In the information extraction stage, the extraction agent can extract structured information from text format data using natural language processing technology. This information may include geographic information, time series information, population changes, and statistical data, which can be used as a data source for the analysis agent. The extracted entity, relationship, and event key information is updated to the public security field knowledge system simultaneously. Structured information can assist subsequent information mining and reasoning, and facilitate the construction of relationships between different entities such as characters and things, and the reasoning of metaphor clues from the structured knowledge system.

[0036] Step 3, analysis agent constructs and analyzes structured information:

[0037] When the analysis agent receives the central control instruction, it calls the structured analysis method to perform information mining based on the expert thinking chain information and the structured information obtained by the extraction agent. Structured analysis mainly affects the reasoning process, strengthens the structural specification of the analysis process, ensures the objectivity and neutrality to ensure the consistency of the evaluation standard, and avoids excluding or affirming a hypothesis or a clue too early, thereby reducing information errors. Applying structured analysis to information mining can make related work more in-depth and refined, and the results more accurate and reliable. After data structuring, the analysis agent can perform structured analysis and other in-depth information mining based on structured information. This may include statistical analysis, trend prediction, association rule mining, clustering analysis, etc. Through these information mining methods, the analysis agent can reveal the patterns and associations behind the data to support decision-making.

[0038] The analysis agent organizes key information such as entities, relationships, and events into structured formats such as tables, charts, and databases. This step usually involves data integration and correlation, unifying data from different sources and formats to form a complete data set, such as analyzing and comparing statistical data over time, generating year-on-year or month-on-month data, etc. According to the statistical data results of the change of geographical position, time sequence, and population change, the statistical data such as geographical information, time sequence information, and population change contained therein are sorted out. Bar chart for data containing different categories or time points, line chart for data containing time-varying data, radar chart for data containing relative size of different variables, trajectory chart for data containing state or position change at different time points, timeline for data containing time sequence and event sequence, pie chart for data containing proportion of each part to the whole, and six types of visualization chart data such as bar chart, line chart, radar chart, trajectory chart, timeline, and pie chart.

[0039] In addition, the analysis agent calls structured analysis methods in the information mining field such as star explosion method, start list method, and quadrant processing method to analyze the content obtained by the search agent, and obtains structured analysis method conclusions.

[0040] Step 4: Writing agent construction and analysis result generation:

[0041] The writing agent generates an analysis report based on the previous structured analysis and visualization analysis results to assist decision-making. The analysis report generated by the writing agent not only depends on the analysis charts generated by the analysis agent, but also depends on the conclusions obtained by the analysis agent based on structured analysis methods such as trend prediction, association rule mining, and clustering analysis. These information and conclusions can help assist decision-making. The suggestions and strategies in the report are based on in-depth data mining and logical reasoning, aiming to help decision-makers clearly understand the nature of the problem when facing complex problems, evaluate different solutions, and choose the best action path.

[0042] In addition, the entire information mining process can also be supplemented by manual intervention for interactive search, analysis, and other information mining operations. The present application simultaneously records the operation logs of user interactive mining and large model agent information mining, which include two parts: behavior action and behavior content. Behavior action includes search, analysis, and writing actions, and behavior content includes search content, search results (including web page title, publishing agency, publishing time, content, URL, etc.), analysis statistical chart data and structured analysis conclusions, and writing analysis report.

[0043] Step 5: Solidification of iterative expert thinking chain template:

[0044] The information mining process can trace back to the information mining process of steps 1 to 4, extract high-value information, and add the behavior action of obtaining the high-value information to the expert thinking chain template.

[0045] If the result of a search, analysis action or behavior in the information mining process is helpful to the final generated analysis report, or part of the generated analysis report is obtained from the search, analysis result, etc., it is considered that the information is high value. According to the behavior content in the behavior log of the intelligent agent and the artificial interactive operation, the high-value information is extracted, and the behavior action of obtaining the high-value information is added to the expert thinking chain template, such as shown in Figure 3 The expert thinking chain template can be iterated once for each information mining.

[0046] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0047] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A high-value information mining method based on an expert thinking chain large model agent, characterized by, Comprise the following steps: Step 1, input the name and summary of the task, extract the entities including persons, time, place, and event information keywords from the large language model, and perform text vectorization processing on the keywords and expert thinking chain templates, match the expert thinking chain template with the highest similarity by calculating the cosine distance similarity between the keywords and the expert thinking chain template vectors; Step 2, call the search engine to search for the entities and event information keywords extracted from the task name and summary, and use the large model and search engine to associate keywords and search related content for each information keyword, get and record the title, publishing agency, publishing time, content, and website of the searched web page, extract the title, publishing structure, and content to get entity, relationship, and event triples, and organize them into a public security domain knowledge system for storage; Step 3, according to the expert thinking chain template, intelligently analyze the constructed public security domain knowledge system to obtain visual statistical charts and structured analysis conclusions; Step 4, according to the visual statistical charts and structured analysis conclusions, write and generate an analysis report to obtain the final information mining analysis report.

2. The high-value information mining method based on the expert thought chain large model agent according to claim 1, characterized in that, Step 1, input the name and summary of the task, extract the entities including persons, time, place, and event information keywords from the large language model, and perform text vectorization processing on the keywords and expert thinking chain templates, match the expert thinking chain template with the highest similarity by calculating the cosine distance similarity between the keywords and the expert thinking chain template vectors, wherein: The expert thinking chain template includes two types of human events and natural disasters, wherein: The human event template is: (1) Please analyze the cause of the event; (2) Please analyze whether the event is a domestic event or an international event; (3) Please analyze the impact of public awareness in the event; (4) Please analyze the strategy and guiding ideology of public security events; (5) Please analyze the hard support of cognitive confrontation in the event; (6) Please analyze the historical background and characteristics of the times in the event; The natural disaster template is: (1) Please analyze the cause of the event; (2) Please analyze the international relations and diplomatic factors of the event; (3) Please analyze the impact of public awareness in the event; (4) Please analyze the strategy and guiding ideology of public security events, disaster emergency plans, government and rescue organization strategies and response measures; (5) Please analyze the hard support of cognitive confrontation in the event; (6) Please analyze the historical background and characteristics of the times in the event.

3. The high-value information mining method based on the expert thought chain large model agent according to claim 1, characterized in that, Step 3, according to the expert thinking chain template, intelligently analyze the constructed public security domain knowledge system to obtain visual statistical charts and structured analysis conclusions, wherein: The intelligent analysis includes two parts of visual data and structured analysis. In the visual data part, column charts are made for data including numerical values of different categories or time points, line charts are made for data including changes over time, radar charts are made for data including relative sizes of different variables, trajectory charts are made for data including state or position changes at different time points, timelines are made for data including time sequence and event sequence, pie charts are made for data including proportions of each part to the whole, and six types of visual statistical charts are made, including column charts, line charts, radar charts, trajectory charts, timelines and pie charts. In the structured analysis part, star explosion method, start list method or quadrant processing method are called to analyze and obtain structured analysis method conclusions.

4. The high-value information mining method based on the expert thought chain large model agent according to claim 1, characterized in that, The operation log is also recorded, which includes two parts of behavior actions and behavior contents. The behavior actions include searching, analyzing and writing, and the behavior contents include search contents, search results, statistical chart data and structured analysis conclusions obtained by analysis, and analysis reports written.

5. The high-value information mining method based on the expert thought chain large model agent according to claim 1, characterized in that, The behavior action of obtaining high-value information is also added to the expert thinking chain template. 6.A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the high-value information mining method based on the expert thinking chain large model agent according to any one of claims 1-5 is implemented, and high-value information mining based on the expert thinking chain large model agent is realized. 7.A computer readable storage medium, having a computer program stored thereon, wherein when the computer program is executed by a processor, the high-value information mining method based on the expert thinking chain large model agent according to any one of claims 1-5 is implemented, and high-value information mining based on the expert thinking chain large model agent is realized.

Images