Social governance-oriented agent non-inductive management method

Abstract

The invention provides a social governance-oriented agent non-inductive management method, which comprises the following steps of: acquiring data related to social events from a plurality of information sources, analyzing the acquired data by utilizing a natural language processing technology and a large language model, and extracting key information related to the social events; subject analysis is carried out on the extracted key information, the key information is converted into a specific governance problem, and the governance problem is split from multiple angles; based on the split governance problem, data related to the current event is retrieved from a plurality of data sources, deep analysis is carried out through expert reasoning, an expert reasoning analysis result is generated, the data sources comprise a pre-constructed knowledge graph, social media and an instant webpage, and expert reasoning adopts a multi-round dialogue reasoning mode; and on the basis of the expert reasoning analysis result, updating a figure portrait, automatically generating a social governance report meeting requirements, and displaying related geographical location information through a geographical information system.

Description

Technical Field

[0001] This application relates to the technical field of social governance, and particularly to an intelligent agent non-intrusive management method for social governance. Background Art

[0002] Currently, the fields of social governance and event analysis mostly rely on traditional methods for problem identification and solution. Traditional technical solutions usually take manual analysis as the core, with information collection relying on manual input, and most information integration processes being relatively scattered, resulting in low timeliness and accuracy of analysis results. These technical solutions have several obvious limitations. Especially when facing complex and rapidly changing social events, they are unable to effectively process a large amount of information and cannot provide accurate decision-making support in a short time.

[0003] The core of social governance lies in identifying and responding to social events, which are usually composed of multiple intertwined factors, including complex factors such as people, locations, event types, and plot developments. Traditional analysis methods rely on experts to evaluate these complex information one by one, and experts judge the influence scope and potential consequences of different events based on experience. However, this process is not only time-consuming but also easily affected by personal experience and biases, resulting in inaccurate decision-making. In actual operation, social governance departments often lack an efficient system to integrate data from different information sources and cannot achieve real-time update and dynamic adjustment of information.

[0004] In terms of information collection, existing technologies usually rely on traditional data collection means, such as news monitoring, public opinion analysis, remote sensing data, etc. These methods often use manual screening and manual classification to process the collected data, with low efficiency and easy omission of important information. For example, although news monitoring can capture a large amount of news content, it is often difficult to automatically identify key elements such as people and geographical information in the articles, and there is a lack of an intelligent analysis framework to guide the accurate extraction of data during the information collection process.

[0005] In addition, traditional social governance analysis technologies often separate event information processing and decision-making, lacking an integrated intelligent platform. Under this architecture, social governance decision-makers must rely on experts from different fields to provide advice, and the opinions of these experts are often isolated and one-sided. When dealing with complex social events, experts often rely on their own domain knowledge for reasoning and analysis, lacking cross-disciplinary comprehensive analysis capabilities. Therefore, the comprehensiveness and depth of event analysis results are greatly limited. More importantly, traditional methods cannot quickly adapt to emergencies, and information updates and adjustments of governance strategies often lag behind the progress of the events themselves.

[0006] In information retrieval, existing technologies usually rely on keyword-matching search engines to extract relevant data. Search engines screen information based on the relevance and authority of web pages, but this text-matching-based retrieval method has significant limitations. Specifically, search engines cannot understand the deep meaning of text, cannot perform semantic analysis on events, and cannot identify complex relationships in text. For example, in news reports involving social events, although the same keywords are used, their contexts and meanings may be completely different, and traditional retrieval methods cannot accurately distinguish these differences, easily leading to interference from irrelevant information.

[0007] Regarding the breakdown of problems, existing technologies usually adopt expert systems or rule-based reasoning methods for analysis, but these methods cannot handle unstructured data and are also difficult to automatically identify potential problems in events. Expert systems reason based on predefined rules, but the rule-making process relies on a large amount of manual input and cannot quickly adapt to new situations. As the complexity of social events increases, the update and maintenance of rules become a huge challenge. At the same time, rule-based systems cannot handle the uncertainty in data, resulting in limitations in the analysis results.

[0008] In summary, existing technologies have significant defects in multiple aspects, especially in data collection, analysis depth, and decision-making efficiency. These drawbacks make existing technologies appear inadequate in dealing with rapidly changing social events. The inefficiency and inaccuracy of traditional methods cannot provide decision-makers with immediate and efficient analysis reports, so there is an urgent need for a technical solution that can automatically process information, provide intelligent analysis, and decision support. Summary of the Invention

[0009] This application aims to at least partly solve one of the technical problems in the related technologies.

[0010] To this end, the first object of this application is to propose an intelligent agent non-intrusive management method for social governance.

[0011] The second object of this application is to propose an intelligent agent workflow.

[0012] The third object of this application is to propose a computer-readable storage medium.

[0013] The fourth object of this application is to propose a computer program product.

[0014] To achieve the above object, the first aspect embodiment of this application proposes an intelligent agent non-intrusive management method for social governance, including:

[0015] Obtain data related to social events from multiple information sources, and use natural language processing techniques and large language models to analyze the obtained data, extract key information related to social events, where the key information includes person information, geographical location information, and event keywords;

[0016] Conduct a theme analysis on the extracted key information, transform it into specific governance issues, and split the governance issues from multiple perspectives;

[0017] Based on the split governance issues, retrieve data related to the current event from multiple data sources and conduct in-depth analysis through expert reasoning to generate an expert reasoning analysis result. The data sources include a pre-constructed knowledge graph, social media, and instant web pages, and the expert reasoning adopts a multi-round dialogue reasoning method;

[0018] Based on the expert reasoning analysis result, update the person portrait, and automatically generate a social governance report that meets the requirements and display the relevant geographical location information through a geographic information system.

[0019] Optionally, the obtaining data related to social events from multiple information sources includes:

[0020] Scrape real-time updated raw data from open-source news websites, social media platforms, and remote sensing data sources. During the scraping process, use a crawler framework for incremental scraping and only obtain the content that has been newly added since the last scrape;

[0021] Clean and integrate the scraped raw data, and use regular expressions to remove the HTML tags and special symbols therein;

[0022] Conduct change detection through remote sensing image data, obtain the detection time, location, and change result, and store the change detection result in JSON format.

[0023] Optionally, the using natural language processing techniques and large language models to analyze the obtained data and extract key information related to social events includes:

[0024] Use natural language processing techniques to perform text preprocessing on the obtained real-time updated data. The preprocessing steps include word segmentation, stop word removal, stemming, and named entity recognition;

[0025] Analyze the preprocessed data through a large language model, identify the core entities and their relationships related to social events in the text, obtain person, geographical location information, and event keywords related to social events, and structure them.

[0026] Optionally, the conducting a theme analysis on the extracted key information, transforming it into specific governance issues, and splitting the governance issues from multiple perspectives includes:

[0027] Conduct topic analysis on the extracted key information through natural language processing and deep learning technologies, identify the core issues of social events, and transform them into issues from multiple governance perspectives;

[0028] According to the obtained themes and specific issues of social events, generate the most relevant types of population information in combination with pre-designed prompt words, including their identities, backgrounds, and positions;

[0029] Fuse different population information and integrate the prompt words in text form into the large language model, and perform multi-group concurrent dialogue processing.

[0030] Optionally, it further includes:

[0031] Judge the rationality of problem splitting through expert feedback or automatically judge the rationality of problem splitting by training a separate model, and automatically adjust the splitting direction and method when needed.

[0032] Optionally, based on the split governance issues, retrieve data related to the current event from multiple data sources, including:

[0033] Vectorize the problem text through a text embedding model to obtain a problem text encoding vector;

[0034] Pre-vectorize the text information of the knowledge graph and social media, and save the vectorization results as files to serve as an auxiliary information vector library;

[0035] Perform matrix operations on the problem text encoding vector and the auxiliary information vector library, and calculate the cosine similarity between the problem text encoding vector and each piece of data in the auxiliary information vector library;

[0036] Take the auxiliary information vectors whose cosine similarity meets the preset requirements as the retrieval results related to the current governance issues.

[0037] Optionally, the process of in-depth analysis through expert reasoning based on the retrieval results includes:

[0038] Based on the retrieval results, start the reasoning process through pre-designed prompt words, provide an initial analysis framework and direction, ensure that the reasoning analysis can focus on the issues of social governance and meet the actual needs;

[0039] During the expert reasoning process, adopt multi-round dialogue reasoning technology based on the Transformer architecture for reasoning. Each round of reasoning is based on the results of the previous round for in-depth exploration and adjustment, gradually optimizing and refining the final solution. Among them, in each round of dialogue, adjust the angle of problem decomposition according to the prompts input by experts, historical reasoning results, and real-time external data, and provide more refined analysis.

[0040] Optionally, based on the expert reasoning analysis results, update the portrait of the person, automatically generate a social governance report that meets the requirements, and display relevant geographical location information through a geographic information system, including:

[0041] Extract information related to the person based on the expert reasoning analysis results, and update the existing portrait of the person, including updating the person's identity, role changes, social relationships, and their behavior patterns in specific events; wherein, the large language model determines whether to adjust the attributes of the portrait of the person by comparing the newly extracted information with the existing portrait database of the person, and automatically maintains and updates based on the information confidence level;

[0042] Extract information related to the geographical location based on the expert reasoning analysis results, and visually display the relevant locations through a geographic information system;

[0043] Based on the expert reasoning analysis results, automatically generate a social governance report that meets the needs of different audiences through natural language generation technology and perform formatting processing; the social governance report includes an overview of the event, analysis of governance issues, social impact assessment, and corresponding policy recommendations to meet the needs of different scenarios.

[0044] To achieve the above object, an embodiment of the second aspect of the present application proposes an intelligent agent workflow, including: a processor module, and a workflow communicatively connected to the processor module;

[0045] The intelligent agent workflow sequentially runs each module according to the design, and performs subsequent task execution and module calls according to the module output results;

[0046] The processor module works according to the working state of the intelligent agent workflow and the pre-designed computer instructions to implement the method described in any one of the above first aspects.

[0047] To achieve the above object, an embodiment of the third aspect of the present application proposes a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in any one of the above first aspects.

[0048] To achieve the above object, an embodiment of the fourth aspect of the present application proposes a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the method described in any one of the above first aspects.

[0049] The technical solutions provided by the embodiments of the present application at least bring the following beneficial effects:

[0050] 1. Intelligent Social Governance: This application combines the powerful advantages of deep learning technology and natural language processing technology, enabling it to automatically extract key information from massive and multi-source data and quickly identify the core issues of social events. Through in-depth understanding and analysis of events, it can provide targeted solutions for social governance. This intelligent processing reduces the reliance on human experts, saves a large amount of time and costs, avoids biases in manual analysis, and improves the efficiency and accuracy of social governance. This application can not only extract key information from traditional texts such as news and reports but also conduct multi-angle analysis through diverse data sources such as social media, providing comprehensive and real-time governance suggestions for decision-makers, thus achieving more efficient and scientific social governance.

[0051] 2. Multi-dimensional Problem Analysis: One of the advantages of this application is its multi-dimensional analysis ability, which can comprehensively analyze social events from different perspectives and the positions of different social groups. Different from traditional methods that usually solve problems from a single angle (such as government, law, or economy), this application can comprehensively consider the impacts of social events on different groups and different stakeholders by using multi-modal data and artificial intelligence algorithms. For example, for the urban pollution problem, this application will not only analyze the causes of environmental pollution but also break down the problem in detail from multiple dimensions such as public health, economic development, and social stability. Through this multi-dimensional analysis, this application can ensure the breadth and depth of problem analysis, provide a more comprehensive and accurate governance perspective, and offer more operable suggestions for decision-makers.

[0052] 3. Automatic Information Update and Reasoning: In a rapidly changing social environment, the accuracy and timeliness of decision-making are crucial. This application has a powerful ability of automatic information update and reasoning, which can process and analyze input data in real time. It can not only continuously update the knowledge base within the system but also automatically update the reasoning results according to new information, ensuring that all governance suggestions are based on the latest data. This means that when social events change, this application can respond quickly, automatically adjust the analysis framework and strategies, thus reducing the risk of information lag. The automatic update mechanism also means that decision-makers can always obtain timely and accurate support, avoiding information update delays caused by manual intervention in traditional methods and significantly improving the governance effect.

[0053] 4. Scalability and Adaptability: The design of this application has a high degree of scalability and adaptability, and can be flexibly configured according to social events of different scales and types. When facing large-scale complex events (such as a nationwide epidemic outbreak or economic crisis), this application can efficiently integrate massive amounts of data and provide comprehensive analysis and governance solutions. When dealing with smaller-scale local events, the system can still provide refined support. In addition, with the continuous development of technology, this application can continuously adapt to new requirements in the field of social governance. For example, in the future intelligent management of cities, this application can combine more Internet of Things (IoT) data and real-time sensor data to further enhance its decision-making support capabilities. The scalability and adaptability of this application not only make it applicable to different types of events, but also can continuously improve its analysis capabilities and reasoning accuracy with the progress of social technology and data processing technology.

[0054] 5. Real-time Feedback and Quick Decision-making Support: This application has a powerful real-time feedback ability, and can quickly obtain relevant information and start analysis at the same time when an event occurs, so as to support decision-makers to take response measures in the shortest time. The system integrates information from multiple channels such as news, social media, and remote sensing data to ensure that decision-makers always have the latest event dynamics. When a social event changes, the system can immediately feedback the adjusted analysis results and propose response strategies. This process greatly shortens the decision-making cycle and improves the speed and accuracy of dealing with emergencies. In the fields of crisis management and emergency response, quick decision-making support can effectively reduce the negative impact of events and improve the response ability of social governance.

[0055] Additional aspects and advantages of this application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of this application. Brief Description of the Drawings

[0056] The above and / or additional aspects and advantages of this application will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0057] Figure 1 is a schematic flow chart of an intelligent agent non-intrusive management method for social governance provided by an embodiment of this application;

[0058] Figure 2 is a schematic structural diagram of an intelligent agent non-intrusive management device for social governance provided by an embodiment of this application. Detailed Embodiments

[0059] Embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, but should not be construed as limiting the present application.

[0060] Referring to the background art, it can be known that the prior art has the following disadvantages:

[0061] 1. The information integration efficiency is low and key information is easily omitted.

[0062] Existing social governance systems usually rely on manual screening and sorting of data, and the information integration process is very cumbersome and inefficient. In this process, data needs to be first extracted from a large number of information sources, then classified and sorted, and finally analyzed. Due to the diversity and complexity of information sources, it is difficult for traditional methods to achieve comprehensive integration and often can only process part of the information, resulting in the omission of important data. For example, geographical information, people's dynamics, public opinion changes, etc. involved in social events are often scattered in different databases and information sources, and the efficiency of manual sorting far cannot meet the rapid response requirements after the occurrence of events. Therefore, a unified and efficient information processing platform cannot be formed to cope with the rapidly developing social events.

[0063] In addition, the prior art also faces challenges in processing cross-domain data. The formats, structures, and qualities of different data sources (such as news, remote sensing information, social media, etc.) vary greatly, which makes it a huge challenge to effectively integrate these heterogeneous data into a unified view. Traditional technologies often rely on manual processing and conversion of different types of data, which is time-consuming and inaccurate, resulting in the quality of information integration cannot be guaranteed.

[0064] 2. The dependence on manual analysis is strong and the results are subjective.

[0065] The prior art generally has the problem of relying on manual analysis. The analysis of social events is usually judged by experts, who often rely on their own experience and knowledge in a certain field to evaluate events. However, the analysis of experts is often strongly subjective and may be affected by factors such as personal biases and cognitive limitations, thus affecting the accuracy and objectivity of the analysis results. Especially when facing complex and cross-domain social events, experts in a single field are difficult to comprehensively and objectively evaluate all dimensions of the events.

[0066] Another disadvantage of manual analysis is its inefficiency. When social events occur, the scope of their impact is usually wide and involves multiple social groups. Expert analysis often requires integrating information and making inferences across multiple fields, which is not only time-consuming but also prone to omissions or misinterpretations due to the large amount of information. As a result, the efficiency of social governance is often limited by the time and capabilities of experts and cannot meet the need to quickly respond to complex events.

[0067] Poor real-time performance and inability to quickly respond to emergencies.

[0068] In existing technologies, the response to social events usually relies on manual judgment and processing, resulting in delays in decision-making and action. Due to the lack of intelligent analysis and automated data collection, existing technologies cannot monitor social events in real time and respond quickly. For example, when emergencies (such as natural disasters, public health events, etc.) occur, traditional methods cannot accurately obtain key information and conduct in-depth analysis in a timely manner, and it often takes several hours or even days to form preliminary decision-making suggestions. Meanwhile, as the event progresses, information changes rapidly, and traditional systems cannot quickly update their data and analysis results. Therefore, decision-makers often face outdated or incomplete analysis information.

[0069] This shortcoming makes traditional technologies appear inadequate in dealing with emergencies. In modern society, emergencies occur at an increasingly fast pace, and the response time of traditional methods cannot meet the requirements of social governance departments for timely response to events, resulting in potential risks not being resolved in a timely and effective manner.

[0070] Insufficient semantic understanding and reasoning capabilities, making it difficult to achieve in-depth analysis.

[0071] In existing technologies, data analysis and reasoning mainly rely on rule-based reasoning or simple statistical analysis methods. These methods may be effective when dealing with simple, structured data, but they show obvious deficiencies when dealing with complex, unstructured data. For example, information such as news reports, social media discussions, and remote sensing images often has a high degree of ambiguity and polysemy, and traditional analysis methods cannot effectively extract and understand the deep meaning in the data. This makes it often impossible for existing technologies to deeply explore potential social problems or response strategies when analyzing social events.

[0072] In addition, the reasoning ability in the existing technologies is relatively single, usually relying on manual rules or simple expert opinions, and unable to conduct multi-dimensional and multi-angle reasoning analysis. The complexity of social governance issues requires taking into account the needs of different positions and different social groups, while the existing technologies are difficult to achieve this. For example, in the analysis of social conflicts, the positions, demands, and potential conflict points of different groups need to be comprehensively considered from multiple dimensions, which is usually impossible to handle for traditional analysis methods.

[0073] 5. Difficult to handle the dynamic changes of large-scale data and information.

[0074] With the acceleration of the informatization process, the amount of data and the change speed of information in social events both show exponential growth. When dealing with large-scale data, the existing technologies usually adopt traditional database and storage methods, and are unable to efficiently store, retrieve, and analyze massive data. For example, when dealing with social events on a global scale, the query and processing speed of traditional technologies is usually slow, unable to meet the requirements of decision-making in a rapidly changing social environment.

[0075] At the same time, the existing technologies are also difficult to cope with the dynamic changes of information. During the analysis process of social events, as the events continue to develop, new information sources will continuously emerge. The existing technologies have poor update and adaptation capabilities, resulting in the analysis results of the system being prone to obsolescence or distortion. Therefore, the existing technologies cannot effectively achieve real-time tracking and dynamic analysis of social events, and decision-makers often face the risk of information lag or inaccuracy.

[0076] 6. Unable to achieve intelligent knowledge update and maintenance.

[0077] In traditional social governance systems, the update and maintenance of knowledge usually rely on manual input and update. This method is not only inefficient but also easily leads to the information in the knowledge base being outdated or inaccurate. As social events continue to develop, the data and rules stored in the knowledge base also need to be continuously updated to adapt to new situations. However, the knowledge update process in the existing technologies often lacks an intelligent mechanism and is unable to automatically adjust the analysis model and the content of the knowledge base according to the development of events. This means that the system cannot achieve continuous learning and adaptation in a dynamic environment, resulting in the system being in a "static" state for a long time and unable to provide efficient support in a complex and changing social environment.

[0078] Aiming at the technical problems and disadvantages existing in the existing technologies, the embodiments of the present application provide an intelligent agent non-intrusive management method for social governance. Specifically, the present application aims to solve the problems such as low information integration efficiency, strong dependence on manual analysis, and slow response speed in traditional methods, so as to provide a more efficient and intelligent social governance solution.

[0079] It should be noted that in some embodiments, in order to better demonstrate the technical solution of the present application, the present application will use virtual modules to explain the functions. The relevant functional modules and their working principles will be Figure 2 shown in detail, and in this way, the implementation process and its advantages of the method of the present application can be more clearly understood.

[0080] As Figure 1 shown, the method includes the following steps:

[0081] Step 101, obtain data related to social events from multiple information sources, and use natural language processing technology and large language models to analyze the obtained data, and extract key information related to social events.

[0082] According to the embodiments of the present application, step 101 involves obtaining data related to social events from multiple information sources, and using natural language processing technology and large language models (LLMs) to analyze these data, and extracting key information therefrom. The key information includes person information, geographical location information, and event keywords. This process involves the functional descriptions of the information source module and the information extraction module.

[0083] In the embodiments of the present application, the information source module is responsible for collecting event-related information from multiple data sources and providing data support for subsequent processing modules. The core task of this module is to capture, clean, and integrate data from different information sources (such as open-source news, remote sensing data, social media, etc.), and provide multi-dimensional information for the present application.

[0084] During the capture process, this module uses an open-source crawler framework to obtain open-source news and social media information in an incremental capture manner with the permission of the other party's website. At the same time, it effectively ensures the timeliness and currency of the data, only capturing the newly added content since the last capture, which helps to save storage space and improve processing efficiency. The captured raw data will be cleaned and integrated. The system uses regular expressions to remove HTML tags and special symbols to ensure the cleanliness and structuring of the captured data. The cleaned data will provide high-quality input data for subsequent analysis.

[0085] In addition, remote sensing image data is another important information source. The input of remote sensing image information is in the form of json text after change detection. It includes the detection time, the longitude and latitude of the detection location, and the results of change detection. Change detection is the process of quantitatively analyzing and determining the characteristics and processes of surface changes from remote sensing data at different times; remote sensing change detection is a process of determining and evaluating various surface phenomena that change over time; remote sensing change detection is the change in the spectral response of image pixels in two periods caused by the change of surface features in the instantaneous field of view of remote sensing over time.

[0086] Next, the information extraction module uses natural language processing techniques and deep learning models to extract key information from the crawled data, such as people, geographical location information, event keywords, etc., and structures the data for subsequent analysis.

[0087] Specifically, the information extraction module first uses natural language processing technology (NLP) for data preprocessing. Specifically, the system performs text preprocessing on the real-time updated data, including steps such as word segmentation, stop word removal, stemming, and named entity recognition. Word segmentation is to split the text into individual words or phrases. Removing stop words can reduce the interference of meaningless common words on the analysis. Stemming helps to unify words with different word forms, and named entity recognition is to extract key entities from the text, such as people, places, organizations, etc.

[0088] Next, the preprocessed data will be input into a large language model (LLM) for in-depth analysis. Based on the powerful text understanding and processing capabilities of LLM technology, LLM has strong text understanding capabilities. Through in-depth analysis of the text, it can identify the core entities related to social events and the relationships between them. These core entities include people, geographical location information, and event keywords. LLM extracts and structures them for subsequent topic extraction and analysis of social governance issues. In this application, through delicate prompt design, LLM can effectively extract information such as people, geographical locations, and event keywords involved in the information provided by the information source module and format them. LLM is a type of language model composed of artificial neural networks with a large number of parameters (usually billions of weights or more), generally based on the transformer architecture, and uses self-supervised learning or semi-supervised learning to train on a large amount of unlabeled text.

[0089] In a possible embodiment, the LLM used in this application is the open-source deepseek-R1 series.

[0090] In addition, it should be noted that in the embodiments of this application, there are also two tools working jointly with the information extraction module: the person information acquisition tool and the geographical information acquisition tool.

[0091] According to the information extracted by the information extraction module, if it involves relevant persons, this application will automatically call the person information acquisition service. The person information acquisition tool will determine whether there are relevant persons in the person portrait library. If there are relevant persons in the person portrait library, the person information acquisition tool will return the relevant person information to the original text of the information extraction module to make the information therein more complete. If the information extracted by the information extraction module involves geographical locations, the geographical information acquisition tool will be automatically called to obtain the geographical location information and geographical particularities of the relevant locations through the Geographic Information System (GIS). If there are no persons and geographical locations involved, no tool will be selected for invocation.

[0092] Through this series of data acquisition and analysis steps, the embodiments of this application can extract key information related to social events from multi-source information, and conduct precise analysis through large language models, providing basic data for subsequent event analysis, problem decomposition, and decision-making support. This process ensures the comprehensiveness, accuracy, and timeliness of information, laying a solid foundation for the effective operation of the intelligent social governance system.

[0093] Step 102: Conduct a theme analysis on the extracted key information, transform it into specific governance problems, and decompose the governance problems from multiple perspectives.

[0094] According to the embodiments of this application, step 102 involves conducting a theme analysis on the key information extracted from multiple information sources through natural language processing and deep learning technologies, and transforming it into specific social governance problems. This process includes two key modules: the theme extraction and problem transformation module, and the problem multi-angle decomposition module.

[0095] In the embodiments of this application, the theme extraction and problem transformation module automatically conducts a theme analysis on the key information extracted in step 101 through natural language processing technologies and deep learning models. This process can help this application identify the core problems of social events and transform them into problems from multiple governance perspectives based on the extracted themes. By designing specific prompts and rules, this application can extract the event themes from the text data provided by the upstream module, and then generate social governance problems related to these themes. These governance problems not only provide a clear framework for subsequent analysis but also can provide targeted solutions for decision-makers.

[0096] The main task of the multi - angle problem - splitting module is to break down social governance problems from multiple dimensions to ensure the comprehensiveness and depth of problem analysis. In this application, the same problem can be analyzed from the perspectives of different social groups, different positions, and social roles. For example, when analyzing a social conflict event, this application can start from the perspectives of different roles such as the government, the public, and enterprises to comprehensively understand all aspects of the problem. This analysis process generates information about several types of people related to the event, including their identities, backgrounds, positions, etc., by combining pre - designed prompt words, and then integrates the different group information into text and inputs it into a large - language model (LLM) for further analysis. This application executes this task through multiple concurrent conversations to ensure in - depth analysis and problem - splitting from multiple dimensions.

[0097] In addition, the embodiment of this application also includes an artificial feedback module, which is used to ensure the rationality of problem - splitting and the accuracy of analysis results. Experts can review the automatically generated analysis results and make adjustments through text feedback to correct the biases in automated analysis. In the process of gradually optimizing this application, the artificial feedback module will gradually be replaced by automated functions, but it still plays a crucial role in the initial stage. Through the collaborative work between experts and this application, the experts' feedback can not only help correct the analysis results but also improve the reasoning logic of the algorithm, enhancing the adaptability and accuracy of this application.

[0098] With the continuous accumulation of experts' feedback, this application can automatically optimize the way and direction of problem - splitting, further improving the quality and efficiency of analysis results. Through the collaborative work between experts' feedback and this application, it not only plays an important role in the problem - splitting stage but also provides data support and optimization basis for the subsequent realization of automated functions.

[0099] In addition, it is also possible to train a model separately to automatically judge the rationality of problem - splitting and automatically adjust the direction and way of splitting when necessary.

[0100] In summary, step 102, through the theme extraction and problem transformation module and the multi - angle problem - splitting module, with the support of the artificial feedback module, enables this application to comprehensively and accurately identify the core problems of social events, break them down and analyze them from multiple angles, providing strong decision - making support for social governance.

[0101] In step 103, based on the split governance problems, retrieve data related to the current event from multiple data sources and conduct in - depth analysis through expert reasoning to generate expert reasoning analysis results.

[0102] According to an embodiment of the present application, the key task of step 103 is to retrieve data related to the current event from multiple data sources based on the decomposed governance issues, and conduct in-depth analysis on the retrieved data through expert reasoning to generate the final expert reasoning analysis result. This process includes two key modules: the information retrieval module and the expert reasoning module.

[0103] In the embodiment of the present application, the information retrieval module mainly retrieves relevant data quickly through the combination of multiple information sources. It includes retrieving data related to the current event from information sources such as pre-constructed knowledge graphs, social media data, and real-time web data, providing more background support for subsequent reasoning and analysis.

[0104] Knowledge graph retrieval stores and manages knowledge in social events through a graph database. In the embodiment of the present application, through semantic queries in the graph database, entities and relationships related to the event can be quickly found. In the knowledge graph, nodes and edges represent entities related to the event (such as people, places, times, etc.) and the relationships between entities.

[0105] In the present application, the semantic retrieval method can be used to calculate the semantic vectors of the questions proposed in the problem multi-angle decomposition module with the entity descriptions and entity relationship descriptions respectively, and select the top n most matching nodes and their relationships. Provide structured data support for the analysis of the subsequent expert module.

[0106] The principle implementation related to all semantic retrievals involved in the present application is as follows: First, by using the bge-m3 text embedding model API of the Zhipu Research Institute (any text embedding model can be used, and the present application does not make specific limitations on this, only for illustration), the question text is vectorized. Through this process, the question text is transformed into a vector of a fixed dimension, capturing the deep semantic information of the text. These encoded vectors will provide an accurate semantic representation for subsequent retrievals. Then, the present application vectorizes the text data in multiple information sources, especially the knowledge graph and social media data. In the pre-constructed knowledge graph, entities related to the event (such as people, places, times, etc.) and the relationships between entities are represented as graph nodes and edges. These data are first vectorized through a text embedding model. Social media data is also processed in the same way, and the vectorized text is stored in the database to form an auxiliary information vector library.

[0107] It can be understood that all vectorized data (including entity relationship data in the knowledge graph and social media information) will be saved as files to form an auxiliary information vector library for subsequent retrieval and call. In this way, the stored vector data not only improves the retrieval efficiency but also provides structured data support for subsequent reasoning and analysis.

[0108] When conducting a search, this application performs matrix operations on the encoded vector of the problem text and each piece of data in the auxiliary information vector library, by calculating the cosine similarity between the encoded vector of the problem text and each piece of data in the auxiliary information vector library. The cosine similarity calculation method can measure the similarity between vectors, that is, measure the semantic relevance between the problem text and each piece of data in the information library. Through this process, this application can find the information source most relevant to the current governance problem.

[0109] After the cosine similarity calculation is completed, this application filters out the search results most relevant to the current governance problem according to a preset threshold. These qualified auxiliary information vectors will be used as search results for the subsequent expert module. Through this semantic-based search method, this application can ensure that the retrieved content is not only based on simple text matching, but based on actual semantic matching, providing more accurate data support.

[0110] As a possible implementation, the specific formula process of text vectorization is as follows:

[0111] Assume the input text sequence x = [x1, x2, …, x n , where x i is the i-th token in the text. First, special tokens [CLS] and [SEP] are added to the text sequence, and then the text sequence is converted into a vocabulary index using a tokenizer. Each token x i (including [CLS] and [SEP]) is mapped into a word vector representation e e through a pre-trained word embedding matrix W i , as shown in formula (1).

[0112] e i = W e x i

[0113] Next, position encoding P i is used to retain the order information of the tokens. Each position i corresponds to a position embedding, as shown in formula (2).

[0114] p i = P i (1)

[0115] Then, the word embedding e i and the position encoding p i are added together to obtain the initial representation e′ i of the token, as shown in formula (3). It contains the position and semantic information of the token x i in the sequence:

[0116] e′ i = ei +p i (2)

[0117] Next, the input is processed through stacked Transformer encoder layers. Each layer uses the self-attention mechanism to model the relationships between tokens. The representation e i ' of each token is mapped to query Q, key K, and value V vectors through a linear transformation, as shown in Equation (4):

[0118]

[0119] where W q , W k , W v are pre-trained weight matrices.

[0120] Next, the self-attention mechanism calculates the weighted average of the output of each token with the keys and values of other tokens, as shown in Equation (5).

[0121]

[0122] where d k is the dimension of the key, and QK T calculates the correlation score (attention weight) between tokens.

[0123] After passing through the self-attention mechanism, the output h i is processed through a feed-forward neural network, as shown in Equation (6), where each layer includes an activation function and a normalization operation:

[0124] FeedForward(h i ) = max(0, h i W1 + b1)W2 + b2 (5)

[0125] where W1, W2 are the weights of the fully connected layers, b1, b2 are the bias terms, and the activation function generally uses ReLU.

[0126] After multiple layers of Transformer encoding, the final representation of each token is a context-aware vector t i , which represents the semantics of the token in the entire context.

[0127] When performing information similarity retrieval, the embodiments of the present application use cosine similarity to measure, and the calculation method of the pre-similarity is as shown in Equation (8).

[0128]

[0129] Among them, A and B respectively represent the semantic encodings of the problem text encoding vector and each text in the auxiliary information vector library. Here, j is the serial number in the vector sequence. For example, if the text is encoded into a 1024-dimensional vector, the value range of j is 1 - 1024. Through matrix calculation and sorting, the text encodings with the top n largest cosine similarities are obtained.

[0130] In addition, in the embodiments of this application, the time information in social media data is also synchronously stored and used as an auxiliary indexing means. The time index helps to quickly locate social media information within a specific time range during actual use, ensuring that the latest information related to the current event can be obtained in real time during event analysis. This method improves the information retrieval efficiency and ensures that social media data can support subsequent analysis tasks in an orderly manner.

[0131] For instant search on web pages, this application will call the Tavily API and pass the query parameters to this API. The specific web page information text content returned will be used for subsequent analysis and reasoning. This method can effectively reduce manual intervention, improve the information acquisition efficiency, and ensure the timeliness and accuracy of the retrieved web page information.

[0132] After the information retrieval is completed, this application will, based on the retrieved relevant data, conduct in-depth analysis on the data through the expert reasoning module to generate the expert reasoning analysis result. The expert reasoning module is one of the core modules in the embodiments of this application, responsible for conducting in-depth reasoning analysis on the information collected from various data sources and generating the final social governance analysis report and solution. By integrating the analysis results of different data types (such as news, remote sensing data, social media information, etc.) and combining with the reasoning model, the embodiments of this application can gradually refine the solution that best meets the social governance requirements and provide specific decision-making suggestions during this process.

[0133] In the embodiments of this application, the expert reasoning module starts the reasoning process based on a pre-designed prompt. The prompt plays an important role here. It provides the initial analysis framework and direction for the model, ensuring that the reasoning analysis can focus on the social governance issues and meet the actual needs. The module adopts a multi-round dialogue reasoning method. Each round of reasoning conducts in-depth exploration and adjustment based on the results of the previous round, gradually optimizing and refining the final solution. Each reasoning not only considers the known historical data but also dynamically absorbs new information feedback, thereby providing a more targeted and operable social governance solution for decision-makers.

[0134] In addition, the multi-turn dialogue reasoning technology based on Transformer endows this module with extremely high flexibility and depth. The advantage of the Transformer architecture lies in its ability to capture long-range dependencies between various parts of the input data through the self-attention mechanism, thus enabling a more accurate understanding and reasoning of complex social events. In each turn of the dialogue, the model can adjust the perspective of problem decomposition according to the prompts input by experts, historical reasoning results, and real-time external data, and provide more refined analysis. This reasoning method can interact and deepen repeatedly to ensure that each governance suggestion has been verified and optimized multiple times, thereby reducing the risk of decision-making and improving the effectiveness of governance solutions.

[0135] Through the expert reasoning module, the embodiments of the present application can simulate the decision-making process of experts, combine big data analysis and machine learning capabilities, and provide strong intelligent support for various decisions in the field of social governance. This not only improves the efficiency of the decision-making process but also enhances the scientificity and accuracy of decisions.

[0136] In summary, step 103, through the collaboration of the information retrieval and expert reasoning modules, provides high-quality decision-making support for social governance problems based on in-depth analysis of multi-source data. This process not only relies on traditional rules and algorithms but also dynamically adapts to new information inputs, gradually optimizing and improving the solution, and enhancing governance efficiency and decision-making timeliness.

[0137] Step 104: Update the person profile based on the expert reasoning analysis results, and automatically generate a social governance report that meets the requirements and display the relevant geographical location information through a geographic information system.

[0138] According to the embodiments of the present application, the key tasks of step 104 are to update the person profile based on the expert reasoning analysis results, automatically generate a social governance report that meets the requirements, and at the same time display the relevant geographical location information through a geographic information system. This process includes three key modules: an information update module, a report generation module, and a geographic information acquisition module. These three modules together constitute the auxiliary function module of the present application, aiming to interact with users efficiently and flexibly adjust its functions according to user needs. Each module operates based on LLM technology, using different prompts and tools to complete specific tasks, ensuring real-time data update, accuracy of report generation, and efficient display of geographic information.

[0139] Specifically, according to the analysis results output by the expert reasoning module, the present application first extracts the involved person information and uses this information to update the existing person portraits. This process involves timely updating the identity, role, social relationships of the person and their behavior patterns in specific events. For example, if the expert reasoning analysis results mention that a certain person has undergone a role change or has a new behavior pattern in an event, the present application will determine whether to adjust the attributes of the person portrait by comparing the newly extracted information with the existing person portrait database. To ensure the accuracy and timeliness of the person portrait, the present application also uses large language model (LLM) technology, which can efficiently process the person descriptions in the text, extract key information, and dynamically update the person portrait.

[0140] During the update process, the information update module of the present application will interact with the existing person portrait system, and automatically update the attributes, behavior patterns, etc. of the person by comparing the old and new data. If new key persons are identified or the roles of existing persons are modified during the expert reasoning process, the information update module can automatically feedback this new information to the person knowledge base to ensure that the person portrait library always remains up-to-date and accurate. This process can ensure that the person portrait is consistent with the actual situation during the development of social events, and helps the social governance department to more accurately identify the behavior patterns and role evolutions of key persons.

[0141] The report generation module is used to automatically generate a social governance report that meets the requirements according to the expert reasoning analysis results. The report generation module ensures that the structure of the report content is clear and well-organized by combining the core points and analysis frameworks in the expert reasoning process. Report generation is not simply outputting the analysis results, but automatically generating report forms that meet the needs of different audiences through natural language generation technology (NLG). These reports can include event overviews, analysis of governance issues, social impact assessments, and corresponding policy recommendations, etc., to meet the needs in different scenarios.

[0142] For example, for a specific social event, the present application can generate a detailed event overview, analyze the background and complexity of the social governance issues, evaluate the social impact of the event, and put forward specific policy recommendations according to the nature and progress of the event. The report generation module can generate intuitive and easy-to-understand reports based on the analysis of the expert reasoning module, combined with actual cases and data charts, to ensure that decision-makers can obtain comprehensive social governance analysis results within a limited time.

[0143] In addition to report generation, this application also extracts information related to geographical locations through a geographical information acquisition module and transfers this information to a Geographic Information System (GIS) for further processing and display. Through this module, this application can convert geographical information related to social events (such as specific cities, regions, event occurrence locations, etc.) into specific coordinate and layer data and visualize it through the GIS system.

[0144] For example, when processing events related to disaster monitoring or analysis of social conflict areas, this application can automatically identify the key geographical location information in the expert reasoning and analysis results and map it onto a map, making the geographical elements clearly visible on the map. This process provides decision-makers with auxiliary support in the spatial dimension and helps them make more comprehensive judgments during the event analysis process.

[0145] Through efficient geographical data processing and graphic rendering technologies, the GIS system can update and display in real-time the geographical information output by the expert reasoning module, helping decision-makers identify the evolution process of social events, changes in geographical hotspots, etc., and thus enhancing the accuracy and scientific nature of decision-making.

[0146] Generally speaking, through the collaborative work of the information update module, report generation module, and geographical information acquisition module, the embodiments of this application achieve dynamic updates of the person portrait, automatic generation of reports, and effective display of geographical information. These modules, through the application of large language model (LLM) technology and natural language generation technology, ensure real-time data updates, the accuracy of report generation, and the efficient display of geographical information, providing comprehensive decision-making support for social governance.

[0147] Step 105, the intelligent agent automation workflow, realizes communication between modules and temporary storage of data, and calls the auxiliary tool module according to the output results of each module

[0148] According to the embodiments of this application, step 105 involves using an intelligent agent to realize communication between modules 101 to 104 in the workflow and temporary storage of data, and calling the auxiliary tool module according to the output results of each module. The core of this step lies in ensuring smooth data flow between modules through the coordination and scheduling of the intelligent agent and being able to dynamically call the auxiliary tool module according to actual needs to support the efficient operation of the entire workflow. In the embodiments of this application, step 105 includes the following key modules: the intelligent agent communication and scheduling module, the data temporary storage module, and the auxiliary tool calling module.

[0149] The Agent Communication and Scheduling Module is responsible for coordinating the communication between the modules in Steps 101 to 104. This module realizes asynchronous communication between modules through agent technology, ensuring that data can be efficiently transmitted between different modules. The agent dynamically schedules the execution order of each module according to preset rules and priorities, avoiding data blocking or processing delays. For example, when Step 102 completes the topic analysis, the agent immediately transmits the analysis result to Step 103 for further processing, while monitoring the status of each module to ensure the continuity and stability of the entire workflow.

[0150] The Data Temporary Storage Module provides temporary storage support when transferring data between modules. Since the processing of the modules in Steps 101 to 104 may involve a large amount of intermediate data, the agent will temporarily store this data in a cache or a distributed storage system for subsequent modules to access quickly. This module also has a data cleaning mechanism that can automatically release the storage space after the data is successfully received by the downstream module, ensuring the efficient use of system resources.

[0151] The Auxiliary Tool Invocation Module dynamically invokes the corresponding auxiliary tool modules according to the output results of each module to support specific tasks. For example, when Step 102 completes the problem splitting, the agent will call a data analysis tool or a visualization tool according to the splitting result to help users understand the problem more intuitively. This module interacts through the interface between the agent and the auxiliary tool module, can flexibly select tools according to actual needs, and automatically releases resources after the task is completed.

[0152] In summary, Step 105 ensures the efficient communication and data processing between the modules in 101 to 104 in the workflow through the collaborative work of the Agent Communication and Scheduling Module, the Data Temporary Storage Module, and the Auxiliary Tool Invocation Module. At the same time, it can dynamically call the auxiliary tool module according to actual needs, providing strong support for the stable operation and efficient decision-making of the entire system.

[0153] To implement the above embodiments, the present application also proposes an agent workflow, including: a processor module, and a workflow communicatively connected to the processor module; the agent workflow sequentially runs each module according to the design, and performs subsequent task execution and module invocation according to the module output results; the processor module works according to the working state of the agent workflow and pre-designed computer instructions to implement the method provided in the foregoing embodiments.

[0154] To implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the method provided in the foregoing embodiments.

[0155] To implement the above embodiments, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the method provided by the foregoing embodiments.

[0156] In the present application, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information and other processing operations all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0157] It should be noted that personal information from users should be collected for legal and reasonable purposes and should not be shared or sold outside of these legal uses. In addition, such collection / sharing should be carried out after obtaining the informed consent of the user, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization including authorizing relevant user information before the user uses this function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy policies and procedures.

[0158] The present application anticipates providing embodiments in which users can selectively block the use or access of personal information data. That is, the present disclosure anticipates providing hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, the risk can be minimized by restricting data collection and deleting the data. In addition, when applicable, personal identifiers are removed from such personal information to protect the privacy of users.

[0159] In the description of the foregoing embodiments, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0160] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, "a plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0161] Any process or method description represented in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logic function or process. The scope of the preferred embodiments of the present application includes additional implementations where functions may be executed not in the order shown or discussed, including in a substantially simultaneous manner according to the functions involved or in a reverse order, which should be understood by those skilled in the technical field of the embodiments of the present application.

[0162] The logic and / or steps represented in a flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be embodied specifically in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0163] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0164] Those of ordinary skill in the art can understand that all or part of the steps carried out in implementing the above-described embodiment methods can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0165] In addition, each functional unit in the various embodiments of the present application can be integrated in a processing module, or each unit can exist physically alone, or two or more units can be integrated in one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0166] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

[0167] It should be understood that various forms of the processes shown above can be used, reordering, adding, or deleting steps. For example, the steps recited in the present application can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present application can be achieved. This is not limited herein.

[0168] The above specific embodiments do not constitute a limitation to the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present application shall be included within the protection scope of the present application.

Claims

1. A non-sensing management method for intelligent entities for social governance, characterized in that: include: The overall framework uses an independently designed intelligent agent workflow, which can automatically obtain data related to social events from multiple information sources, and use natural language processing technology and large language models to analyze the acquired data to extract key information related to social events, including character information, geographic location information and event keywords; Conduct thematic analysis on the extracted key information, transform it into specific governance issues and decompose the governance issues from multiple perspectives; Based on the split governance issues, data related to the current event is retrieved from multiple data sources and analyzed in depth through expert reasoning to generate expert reasoning analysis results. The data sources include pre-built knowledge graphs, social media, and instant web pages. The expert reasoning adopts a multi-round dialogue reasoning method. Based on the expert reasoning and analysis results, the character portrait is updated, and a social governance report that meets the needs is automatically generated and the relevant geographic location information is displayed through the geographic information system.

2. The method according to claim 1, characterized in that The data related to social events are obtained from multiple information sources, including: Crawl real-time updated raw data from open source news sites, social media platforms, and remote sensing data sources, using a crawler framework to perform incremental crawling during the crawling process, only obtaining new content since the last crawl; Clean and integrate the captured raw data, and use regular expressions to remove HTML tags and special symbols; Change detection is performed through remote sensing image data to obtain the detection time, location and change results, and the change detection results are stored in JSON format.

3. The method according to claim 2, characterized in that The natural language processing technology and large language model are used to analyze the acquired data and extract key information related to social events, including: Using natural language processing technology to perform text preprocessing on the acquired real-time update data, the preprocessing steps include word segmentation, stop word removal, stem extraction and named entity recognition; The preprocessed data is analyzed through a large language model to identify the core entities and their relationships related to social events in the text, obtain the people, geographic location information and event keywords related to the social events, and structure them.

4. The method according to claim 3, characterized in that The extracted key information is subjected to thematic analysis, converted into specific governance issues, and the governance issues are broken down from multiple perspectives, including: Conduct thematic analysis of extracted key information through natural language processing and deep learning technologies, identify the core issues of social events, and transform them into issues from multiple governance perspectives; Based on the themes and specific issues of the social events obtained, combined with pre-designed prompt words, the most relevant information about several groups of people is generated, including their identities, backgrounds, and positions; The information of different groups is integrated into the large language model in the form of text, and multiple groups of concurrent dialogues are processed.

5. The method according to claim 4, characterized in that Also includes: The rationality of problem splitting can be judged through expert feedback or automatically judged through separate training models, and the direction and method of splitting can be automatically adjusted when necessary.

6. The method according to claim 5, characterized in that Based on the split governance issues, data related to the current event is retrieved from multiple data sources, including: Vectorize the question text through the text embedding model to obtain the question text encoding vector; Pre-vectorize the text information of the knowledge graph and social media, and save the vectorization results as files as auxiliary information vector libraries; Perform matrix operations on the question text encoding vector and the auxiliary information vector library, and calculate the cosine similarity between the question text encoding vector and each piece of data in the auxiliary information vector library; The auxiliary information vector whose cosine similarity meets the preset requirements is taken as the retrieval result related to the current governance issue.

7. The method according to claim 6, characterized in that The process of conducting in-depth analysis based on search results through expert reasoning includes: Based on the search results, the reasoning process is initiated through pre-designed prompt words, providing an initial analysis framework and direction to ensure that the reasoning analysis can be carried out around the issues of social governance and meet actual needs; In the expert reasoning process, multi-round dialogue reasoning technology based on the Transformer architecture is used for reasoning. Each round of reasoning conducts in-depth exploration and adjustment based on the results of the previous round, and gradually optimizes and refines the final solution. In each round of dialogue, the angle of problem decomposition is adjusted according to the prompts input by the expert, historical reasoning results and real-time external data, and a more detailed analysis is provided.

8. The method according to claim 7, characterized in that Based on the expert reasoning and analysis results, the character portrait is updated, and a social governance report that meets the needs is automatically generated and the relevant geographic location information is displayed through the geographic information system, including: Based on the expert reasoning and analysis results, information related to the person is extracted, and the existing person portrait is updated, including updating the person's identity, role changes, social relations, and behavior patterns in specific events; wherein the large language model compares the newly extracted information with the existing person portrait database to determine whether the attributes of the person portrait need to be adjusted, and automatically maintains and updates it based on the information confidence; Based on the expert reasoning and analysis results, extract information related to the geographic location and visualize the relevant locations through a geographic information system; Based on the results of the expert reasoning and analysis, a social governance report that meets the needs of different audiences is automatically generated and formatted through natural language generation technology; the social governance report includes an event overview, governance issue analysis, social impact assessment and corresponding policy recommendations to meet the needs of different scenarios.

9. An agent workflow, characterized in that: include: A processor module, and a workflow in communication with the processor module; The agent workflow runs each module in sequence according to the design, and performs subsequent tasks and module calls according to the module output results; The processor module operates according to the working state of the agent workflow and pre-designed computer instructions to implement the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

Images