Information management method and system based on multi-agent collaborative decision-making mechanism

Abstract

The invention discloses an information management method and system based on a multi-agent collaborative decision-making mechanism. The method comprises the following steps: constructing a large-model multi-agent system which comprises a policy interpretation agent, a data acquisition agent and an index analysis agent; collecting multi-modal data input by a user by using a data collection agent, wherein the multi-modal data comprises text data and non-text data; performing quantification processing on the text data and the non-text data; outputting a data authenticity pre-evaluation result by using the data acquisition agent; outputting a compliance pre-evaluation result by using the policy interpretation agent; outputting an advancement pre-evaluation result by using the index analysis agent; dynamically adjusting the weight value of each agent according to different evaluation scenes; and generating an information evaluation result according to the weight value of each agent and the evaluation result of each agent. According to the method, by constructing schemes such as multi-modal data quantification and dynamic weight adjustment of each agent, quantifiable and collaborative operation of information management is realized.

Description

Technical Field

[0001] This application relates to the field of data information management technology, and more specifically, to an information management method and system based on a multi-agent collaborative decision-making mechanism. Background Technology

[0002] With the continuous development of society, promoting organizational work on the internet is a current trend. However, existing information management systems suffer from limited application scenarios and a lack of systematic collaborative capabilities. For example, some systems can only perform single tasks such as learning updates and activity notifications, without building specialized intelligent agent divisions of labor. A few systems with basic collaborative capabilities only achieve simple data transfer through fixed permission assignments, failing to simulate the full-process collaborative logic of "policy interpretation - task decomposition - execution feedback" in real-world scenarios, and the matching degree between intelligent agent roles and management responsibilities is insufficient. Furthermore, existing evaluation methods still rely on traditional models. Data processing is limited to structured data (such as learning duration and activity participation frequency). There is a lack of effective quantification methods for unstructured data such as ideological reports, organizational life audio, and activity scene images. Evaluation models mainly rely on "manually set indicator weights + simple statistics," failing to introduce multimodal technology to accurately extract evaluation features or establish an automatic matching mechanism with regulations and clauses. This results in highly subjective evaluation results, a lack of compliance verification, and difficulty in supporting the standardization and precision of advanced judgments. Summary of the Invention

[0003] The main purpose of this application is to provide an information management method and system based on a multi-agent collaborative decision-making mechanism to solve the problems of insufficient matching between agent roles and management responsibilities and lack of effective quantification methods for unstructured data in the prior art.

[0004] One embodiment of this application provides an information management method based on a multi-agent collaborative decision-making mechanism, including the following steps:

[0005] Construct a large-scale model multi-agent system, which includes a policy interpretation agent, a data collection agent, and an indicator analysis agent;

[0006] The data acquisition agent is used to collect multimodal data input by the user, including text data and non-text data;

[0007] The text data is quantified by extracting policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics, and calculating the text quantification value based on the policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics; the non-text data is also quantified by extracting activity participation characteristics, sentiment characteristics, and public satisfaction characteristics, and calculating the non-text quantification value based on the activity participation characteristics, sentiment characteristics, and public satisfaction characteristics.

[0008] The data collection agent outputs a preliminary assessment result on the authenticity of the data; the policy interpretation agent outputs a preliminary assessment result on compliance; and the indicator analysis agent outputs a preliminary assessment result on the advancement of the technology.

[0009] The weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent are dynamically adjusted according to different evaluation scenarios.

[0010] Information evaluation results are generated based on the weight values ​​of the policy interpretation intelligence agent, the data collection intelligence agent, the indicator analysis intelligence agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results.

[0011] In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: comparing text data with core clauses of regulations to obtain the semantic similarity between the text data and the core clauses of regulations; obtaining work performance keywords in the text data and calculating their TF-IDF weighted values ​​based on the work performance keywords; extracting the quantitative values ​​of problem rectification entities in the text data using named entity recognition technology; and calculating the text quantification value based on the semantic similarity, the TF-IDF weighted values ​​of the work performance keywords, and the quantitative values ​​of the problem rectification entities.

[0012] In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: dividing non-text data into image data, voice data, and public evaluation text; identifying the proportion of actual participants to expected participants in the image data using image detection technology; calculating the proportion of positive emotions in the voice data of the organizational life meeting using voice sentiment analysis technology; calculating the proportion of positive evaluations in the public evaluation text using text sentiment analysis technology; and calculating the non-text quantification value based on the proportion of actual participants to expected participants, the proportion of positive emotions in the voice data of the organizational life meeting, and the proportion of positive evaluations in the public evaluation text.

[0013] In some embodiments, the process of dynamically adjusting the weight values ​​of the policy interpretation intelligence, the data collection intelligence, and the indicator analysis intelligence according to different evaluation scenarios includes the following steps: setting the initial weights of the policy interpretation intelligence, the data collection intelligence, and the indicator analysis intelligence; in one evaluation scenario, calculating the weight of the policy interpretation intelligence based on the policy fit characteristics, the initial weight of the policy interpretation intelligence, and a first weight adjustment coefficient; calculating the weight of the data collection intelligence based on the activity participation characteristics, the initial weight of the data collection intelligence, and a second weight adjustment coefficient; and so on. The weight of the indicator analysis agent is calculated based on the work performance characteristics, the initial weight of the indicator analysis agent, and the third weight adjustment coefficient. In another evaluation scenario, the weight of the policy interpretation agent is calculated based on the policy fit characteristics, the initial weight of the policy interpretation agent, and the first weight adjustment coefficient. The weight of the data collection agent is calculated based on the public satisfaction characteristics, the initial weight of the data collection agent, and the fourth weight adjustment coefficient. The weight of the indicator analysis agent is calculated based on the performance and dedication characteristics, the initial weight of the indicator analysis agent, and the fifth weight adjustment coefficient. Among these, the performance and dedication characteristics are extracted from volunteer activity records by the large model.

[0014] In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: in one evaluation scenario, under the premise of meeting the discipline and compliance dimension, the first weight adjustment coefficient is automatically increased to the sixth weight adjustment coefficient to increase the weight of the policy interpretation agent; in another evaluation scenario, when the large model detects a tendency for learning and education to fall short of standards, the first weight adjustment coefficient is automatically increased to the seventh weight adjustment coefficient to increase the weight of the policy interpretation agent.

[0015] In some embodiments, the large model multi-agent system further includes an expert interactive agent, which generates correction coefficients to correct the information evaluation results.

[0016] In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: obtaining expert opinions through the expert interactive agent; calculating the fit between the expert opinions and the scenario to be evaluated based on semantic analysis technology; determining the weight coefficient of the expert opinions based on the expert's years of service and richness of evaluation experience; generating the correction coefficient based on the fit between the expert opinions and the scenario to be evaluated and the weight coefficient of the expert opinions; and determining the corrected information evaluation result based on the correction coefficient and the information evaluation result.

[0017] In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: setting a reward signal, wherein the reward signal is derived by combining the degree of consistency between the corrected information evaluation result and the subsequent actual performance result after a preset time and the proportion of the number of expert corrections to the total number of evaluations; and iterating on any one or more of the first weight adjustment coefficient, the second weight adjustment coefficient, the third weight adjustment coefficient, the fourth weight adjustment coefficient, and the fifth weight adjustment coefficient according to the reward signal.

[0018] Another embodiment of this application provides an information management system based on a multi-agent collaborative decision-making mechanism, including:

[0019] The large-scale model multi-agent module includes a policy interpretation agent, a data collection agent, and an indicator analysis agent.

[0020] The data acquisition agent is used to collect multimodal data input by users and to quantify the text and non-text data within the multimodal data. During the quantification of text data, policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics are extracted, and a text quantification value is calculated based on these characteristics. During the quantification of non-text data, activity participation characteristics, sentiment characteristics, and public satisfaction characteristics are extracted, and a non-text quantification value is calculated based on these characteristics. The data acquisition agent outputs a pre-assessment result of data authenticity based on the data consistency and integrity of the multimodal data. The policy interpretation agent outputs a pre-assessment result of compliance based on the policy fit characteristics and the number of matching compliance clauses. The indicator analysis agent outputs a pre-assessment result of advancement based on the analytical indicators of the assessment scenario.

[0021] The dynamic weight decision module is used to dynamically adjust the weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent according to different evaluation scenarios.

[0022] The multi-agent collaborative decision-making module is used to generate information evaluation results based on the weight values ​​of the policy interpretation agent, the data collection agent, the indicator analysis agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results.

[0023] In some embodiments, the large model multi-agent system further includes an expert interactive agent, which generates correction coefficients to correct the information evaluation results;

[0024] The multi-agent collaborative decision-making module includes: a large-scale model joint pre-analysis sub-module, used to generate information evaluation results based on the weight values ​​of the policy interpretation agent, the data collection agent, the indicator analysis agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results; an expert correction quantification sub-module, used to generate correction coefficients to correct the information evaluation results; a decision result output and application sub-module, used to generate a standardized evaluation report adapted to multiple terminals from the corrected information evaluation results; and a full-link traceability and responsibility definition sub-module, used to record key information of each link to form a traceable link.

[0025] The information management method and system based on a multi-agent collaborative decision-making mechanism proposed in this application have the following advantages and beneficial effects:

[0026] 1. In the information management method based on a multi-agent collaborative decision-making mechanism provided in this application, by setting up a policy interpretation agent, a data collection agent, and an indicator analysis agent, and enabling these agents to operate collaboratively, the information management method provided in this application becomes more adaptable to hierarchical needs. Existing management systems often employ single agents or fixed-permission collaboration, failing to achieve multi-agent collaborative operation and adaptability to regulatory priority requirements. This application, through a regulatory-anchored multi-agent division of labor and dynamic weight adjustment model, achieves policy impartiality and execution adaptation in collaborative decision-making. In practical applications, this mechanism can significantly improve the policy compliance rate and task execution adaptability of cross-level collaborative decision-making, greatly reducing manual coordination costs. At the level of specialized division of labor, the policy interpretation agent focuses on regulatory analysis, ensuring that decisions are always based on regulations, such as automatically verifying whether activities align with the core requirements of the service center's work, avoiding formal compliance with substantive deviations. The data collection agent and the indicator analysis agent are respectively responsible for multimodal data quantification and dual-index calculation, forming a specialized collaborative link of "policy oversight - data support - indicator analysis," solving the problems of "ambiguous role division and chaotic collaborative logic" in existing technologies. At the dynamic weighting level, based on dual-index evaluation characteristics (such as policy fit and activity participation), the information management method can adaptively adjust the weights of intelligent agents. For example, when evaluating disciplinary compliance, the weight of the policy interpretation intelligent agent is automatically increased to ensure compliance takes priority. When evaluating performance, the influence of public satisfaction in the data collection intelligent agent is strengthened, taking into account both public approval and policy requirements, thus overcoming the limitations of existing technologies that set fixed weights for each indicator and cannot adapt to different scenarios.

[0027] 2. In the information management method based on a multi-agent collaborative decision-making mechanism provided in this application, a dual-index evaluation feature quantification mechanism driven by a large-scale model multimodal approach is constructed in the data acquisition agent to achieve the quantification of both text and non-text data. In existing information management systems, the evaluation of dual indices (fortress index and vanguard index) largely relies on manual scoring and structured data, such as the number of activities. The utilization rate of unstructured data, such as reports and activity photos, is poor, resulting in highly subjective evaluation results and poor comparability. This application achieves full data coverage and full-process quantification of dual-index evaluation by employing a large-scale model multimodal quantification mechanism and a quantitative calculation model in the data acquisition agent. This mechanism can significantly reduce human error in dual-index evaluation and significantly improve the consistency between evaluation results and actual performance. At the data dimension level, this application introduces large-scale multimodal technology to process three types of unstructured data: text, image, and speech. For example, it extracts policy relevance through text semantic analysis, identifies activity participation rates through image detection, and obtains ideological dynamics through speech sentiment analysis. This expands the evaluation data dimension from a single structured data to a multimodal fusion, addressing the problem of incomplete data dimensions in existing technologies. At the computational logic level, through standardized formulas, such as text quantification values ​​and non-text quantification values, multimodal features are transformed into computable parameters. These parameters are then combined with dynamic weights to generate dual-index pre-evaluation results, upgrading the evaluation results from qualitative descriptions to quantitative scores. Evaluation results from different regions and levels are comparable and traceable.

[0028] 3. In one embodiment, the large-scale model multi-agent system further includes an expert interactive agent, which generates correction coefficients to correct the information evaluation results. As needed, the information management method based on the multi-agent collaborative decision-making mechanism provided in this application can also set a reward signal. The reward signal is derived by combining the degree of agreement between the corrected information evaluation results and the subsequent actual performance results after a preset time, as well as the proportion of expert corrections to the total number of evaluations. Then, based on the reward signal, any one or more of the first weight adjustment coefficient, the second weight adjustment coefficient, the third weight adjustment coefficient, the fourth weight adjustment coefficient, and the fifth weight adjustment coefficient are iterated. The above method realizes human-machine collaborative decision-making and achieves an efficient closed loop in the human-machine collaborative decision-making process. In existing information management systems, the integration of "large-scale model + human" remains at a shallow stage of model output and independent human decision-making. Pre-analysis lacks quantitative support, feedback lacks an iterative mechanism, decision-making efficiency is low, and deviations repeatedly occur. This application achieves efficient integration and continuous optimization of human-machine collaboration through quantitative collaborative processes and reinforcement learning iterations. In one embodiment, the information management system also includes a full-link traceability and responsibility delineation submodule, used to record key information at each stage, forming a traceable closed-loop chain. Through model analysis of the entire chain record, the root causes of decision-making deviations (such as multimodal feature extraction errors or human error correction) can be traced, solving the problem of ambiguous human-machine responsibility delineation in existing technologies and improving the credibility of technology applications. Regarding decision-making efficiency, large-scale model joint pre-analysis can automatically integrate multimodal data, verify policy compliance, and generate structured pre-assessment results with accompanying data and compliance reports. Experts do not need to sift through information from scratch; they only need to make corrections for specific scenarios, significantly shortening the time for a single decision and significantly improving efficiency. Regarding decision-making accuracy, expert correction coefficients and reward signals are introduced, transforming expert experience into calculable parameters, such as scenario adaptability. Then, through algorithmic iteration to optimize the weight adjustment coefficients, the expert correction rate in model pre-analysis is significantly reduced, minimizing the recurrence of similar decision-making deviations. Attached Figure Description

[0029] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0030] Figure 1 This is a flowchart illustrating an information management method based on a multi-agent collaborative decision-making mechanism disclosed in one embodiment of this application.

[0031] Figure 2 for Figure 1 A flowchart illustrating the process of quantizing text data in China;

[0032] Figure 3 for Figure 1 A flowchart illustrating the process of quantizing non-text data in China;

[0033] Figure 4 for Figure 1 A flowchart illustrating the process of dynamically adjusting the weights of each agent.

[0034] Figure 5 for Figure 1 A flowchart illustrating another embodiment of dynamically adjusting the weights of each agent.

[0035] Figure 6 for Figure 1 A flowchart illustrating the process of adding a new expert agent in the process;

[0036] Figure 7 for Figure 6 A flowchart illustrating the iterative process for each weight coefficient;

[0037] Figure 8 This is a schematic diagram of a module of an information management system based on a multi-agent collaborative decision-making mechanism disclosed in another embodiment of this application;

[0038] Figure 9 for Figure 8 A schematic diagram of the data acquisition agent module in the diagram;

[0039] Figure 10 for Figure 8 A flowchart illustrating the workflow of an information management system based on a multi-agent collaborative decision-making mechanism.

[0040] Figure 11 This is a schematic diagram of a computer device disclosed in one embodiment of this application. Detailed Implementation

[0041] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0044] Please see Figure 1 One embodiment of this application provides an information management method based on a multi-agent collaborative decision-making mechanism. The management method includes the following steps:

[0045] A large-scale model multi-agent system is constructed, comprising a policy interpretation agent, a data collection agent, and an indicator analysis agent. Specifically, the policy interpretation agent outputs compliance pre-assessment results, the data collection agent outputs data authenticity pre-assessment results, and the indicator analysis agent outputs advancement pre-assessment results.

[0046] The data acquisition agent is used to collect multimodal data input by the user, including text data and non-text data;

[0047] The text data undergoes quantification processing, extracting policy relevance features, work effectiveness features, and problem rectification features, and calculating text quantification values ​​based on these features. Similarly, non-text data undergoes quantification processing, extracting activity participation features, sentiment characteristics, and public satisfaction features, and calculating non-text quantification values ​​based on these features. Specifically, a large-scale, multimodal driven dual-index evaluation feature quantification mechanism is constructed within the data acquisition agent. This dual-index evaluation feature quantification mechanism can quantify both text and non-text data.

[0048] The data collection agent outputs a preliminary assessment result on the authenticity of the data; the policy interpretation agent outputs a preliminary assessment result on compliance; and the indicator analysis agent outputs a preliminary assessment result on the advancement of the technology.

[0049] Based on different evaluation scenarios, the weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent are dynamically adjusted. Specifically, a dual-exponential driven dynamic weight adaptive decision model can be constructed. In this model, the weight values ​​of each agent can be dynamically adjusted according to different evaluation scenarios.

[0050] Information evaluation results are generated based on the weight values ​​of the policy interpretation intelligence agent, the data collection intelligence agent, the indicator analysis intelligence agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results.

[0051] In this embodiment, the large-scale multi-agent system is anchored to regulations. By designing a division-of-labor architecture for policy interpretation agents, data collection agents, and indicator analysis agents, each agent achieves collaborative linkage through standardized data interfaces, ensuring that decisions conform to policy principles and are adapted to the execution scenario.

[0052] The policy interpretation intelligent agent is primarily used for regulation analysis and compliance verification, extracting key policy clauses and building a mapping library. This agent can automatically verify the compliance of targets and indicators during decision-making, providing warnings and corrective suggestions. Specifically, the agent focuses on regulations and the latest policies, building policy analysis and dynamic verification capabilities. It integrates policy content to form a mapping library of bottom-line requirements, obligations, and compliance assessment standards, automatically updating the policy library monthly. In various scenarios, it automatically verifies process compliance. For example, if it detects issues such as failure to conduct required warning education or insufficient training duration for activists (less than one year), the agent will immediately push excerpts of the original policy text and rectification cases to ensure that work always aligns with policy guidance.

[0053] The data collection agent is primarily used for unified access and preliminary quantification of multimodal data, including text, images, and voice. It extracts key features through a large model interface to provide data support for evaluation. In this embodiment, the data collection agent uses a hybrid storage architecture of MySQL and MongoDB for three-dimensional classification and on-demand distribution. Structured data (such as participation rate and satisfaction parameters) is stored in MySQL to ensure computational efficiency. Raw multimodal data (such as thought reports and activity photos) is stored in MongoDB and indexed. The data collection agent can dynamically distribute data according to the needs of other agents. For example, when the indicator analysis agent calculates the fortress index, it automatically retrieves the corresponding activity data and public evaluation data, achieving cross-agent data interoperability and sharing.

[0054] The indicator analysis agent is primarily used to calculate core indicators and eliminate invalid data by combining evaluation models and policy requirements. In this embodiment, the indicator analysis agent is used for the quantitative calculation of the fortress index and the vanguard index. The indicator analysis agent covers the entire process of the five major stages and 25 procedures of the cultivation. When calculating the fortress index, evaluation dimensions are set around organizational construction (frequency of activities, policy implementation), service effectiveness (number of times the public is assisted, duration of volunteer practice), and discipline and compliance (timely payment rate, incidence of violations). When calculating the vanguard index, core indicators are set by combining learning and education (duration of MOOC learning, quality of ideological reports), performance of duties and contributions (job performance, number of volunteer services), and public evaluation (positive review rate). At the same time, data that "meets the standard in form but violates the standard in substance" (such as meeting the activity frequency standard but not serving the central work of rural revitalization) is automatically eliminated to ensure that the evaluation results truly reflect the work effectiveness.

[0055] As needed, the large-scale model multi-agent system also includes an expert interactive agent. This expert interactive agent primarily serves to bridge the gap between pre-analysis and expert judgment, compiling pre-assessment results and pushing them to experts; collecting and transforming expert opinions and feeding them back to the decision-making module. In this embodiment, the expert interactive agent builds a human-machine collaborative bridge between large-scale model pre-analysis and expert fine-tuning, achieving the integration of expert experience and technical analysis. The expert interactive agent automatically integrates the output results of the three types of agents, generating a structured form containing multimodal data (screenshots of policy matching clauses, activity photo participation rate analysis, and public comment segments), compliance verification conclusions, and dual-index pre-scoring, which is then pushed to the experts. Experts can annotate and correct opinions online for specific scenarios, and through large-scale model semantic analysis, transform these opinions into calculable parameters such as scenario adaptability coefficients and indicator weight adjustment values, feeding them back to the dynamic decision-making module to improve the accuracy of assessments in specific scenarios.

[0056] In reality, existing information management systems suffer from a mismatch between multi-agent collaboration and management needs. The core issue lies in the lack of multi-modal fusion and collaboration logic. Lacking a central hub centered on a large model and failing to establish an agent architecture adapted to the hierarchical structure, agents can only process single data types and cannot integrate multi-modal information such as images, voice, and text to achieve automatic task decomposition and adaptation across levels. Furthermore, existing information management systems lack dynamic logic supporting multi-modal scenarios for adjusting decision weights, still relying on fixed permission allocation models. They fail to leverage the large model's deep analysis of regulations and multi-modal scenario characteristics to dynamically adjust agent weights to balance policy compliance and feasibility. In conflict resolution, a multi-modal-driven dialogue and negotiation mechanism is also absent, with multiple levels of conflicting needs still relying on manual coordination, thus failing to unlock the multi-modal collaborative value of the large model.

[0057] In the information management method provided in this application, by setting up a policy interpretation agent, a data collection agent, and an indicator analysis agent, and enabling these agents to operate collaboratively, the information management method provided in this application becomes more adaptable to hierarchical needs. Existing management systems often employ single agents or fixed-permission collaboration, failing to achieve collaborative operation and compliance with regulatory priority requirements. This invention, through a multi-agent division of labor and dynamic weight adjustment model anchored to regulations, achieves policy impartiality and execution adaptation in collaborative decision-making. In practical applications, this mechanism can significantly improve the policy compliance rate and task execution adaptability of cross-level collaborative decision-making, greatly reducing manual coordination costs. At the level of specialized division of labor, the policy interpretation agent focuses on regulation analysis, ensuring that decisions are always based on regulations, such as automatically verifying whether activities align with the core requirements of the service center's work, avoiding formal compliance with substantive deviations. The data collection agent and the indicator analysis agent are respectively responsible for multimodal data quantification and dual-index calculation, forming a specialized collaborative link of "policy oversight - data support - indicator analysis," solving the problems of "ambiguous role division and chaotic collaborative logic" in existing technologies. At the dynamic weighting level, based on dual-index evaluation characteristics (such as policy fit and activity participation), the information management method can adaptively adjust the weights of intelligent agents. For example, when evaluating disciplinary compliance, the weight of the policy interpretation intelligent agent is automatically increased to ensure compliance takes priority. When evaluating performance, the influence of public satisfaction in the data collection intelligent agent is strengthened, taking into account both public approval and policy requirements, thus overcoming the limitations of existing technologies that set fixed weights for each indicator and cannot adapt to different scenarios.

[0058] Furthermore, existing information management systems suffer from a lack of quantification and intelligence in dual-index assessment technology. Current large-scale multimodal technology exhibits a technological gap in dual-index assessment, failing to leverage the advantages of large models in processing unstructured data. For various data types, such as reports on ideological work, audio recordings of organizational activities, and images from event scenes, only superficial processing is possible, failing to achieve cross-modal feature fusion and quantitative transformation, resulting in a lack of calculable parameters for core assessment dimensions. The assessment model does not incorporate multimodal compliance verification, policy interpretation relies on manual intervention, and automatic matching and verification algorithms are lacking. Advancement judgments remain primarily manual, lacking standardized computational logic supported by multimodal approaches, leading to highly subjective assessment results and poor comparability.

[0059] In the information management method provided in this application embodiment, a dual-index evaluation feature quantification mechanism driven by a large-scale model multimodal approach is constructed in the data acquisition agent to achieve the quantification process of text and non-text data. In existing information management systems, the evaluation of dual indices (fortress index and vanguard index) largely relies on manual scoring and structured data, such as the number of activities. The utilization rate of unstructured data, such as ideological reports and activity photos, is poor, resulting in highly subjective evaluation results and poor comparability. This invention achieves full data coverage and full-process quantification of dual-index evaluation through a large-scale model multimodal quantification mechanism and a quantitative calculation model. This mechanism can significantly reduce human error in dual-index evaluation and significantly improve the consistency between evaluation results and actual performance. At the data dimension level, this application introduces large-scale model multimodal technology to process three types of unstructured data: text, image, and voice. For example, policy fit is extracted through text semantic analysis, activity participation rate is identified through image detection, and ideological dynamics are obtained through voice sentiment analysis. The evaluation data dimension is expanded from a single structured data to a multimodal fusion to solve the problem of incomplete data dimensions in the prior art. At the computational logic level, multimodal features are transformed into computable parameters through standardized formulas, such as text quantification values ​​and non-text quantification values. Then, combined with dynamic weights, dual-index pre-evaluation results are generated, upgrading the evaluation results from qualitative descriptions to quantitative scores. Evaluation results from different regions and levels are comparable and traceable.

[0060] Please see also Figure 2 In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: comparing text data with the core clauses of regulations to obtain the semantic similarity between the text data and the core clauses of regulations; obtaining work performance keywords in the text data and calculating their TF-IDF weighted values ​​based on the work performance keywords; extracting the quantitative values ​​of problem rectification entities in the text data using named entity recognition technology; and calculating the text quantification value based on the semantic similarity, the TF-IDF weighted values ​​of the work performance keywords, and the quantitative values ​​of the problem rectification entities.

[0061] Specifically, the text quantization value is calculated in the following way:

[0062] T = α·Sim(T text , T rule ) + β·TF-IDF(T text , K effect ) + γ·NER(T text E rectify );

[0063] Where T is the text quantization value; Sim(T) text , T rule) represents the semantic similarity between the report text and the core clauses of the regulations calculated by the large model; TF-IDF(T) text , K effect ) represents the TF-IDF weighted value of keywords related to work effectiveness in the text; NER(T) text E rectify ) represents the quantitative value of the problem rectification entity extracted by named entity recognition in the large model; α represents the policy fit coefficient; β represents the work effectiveness coefficient; and γ represents the problem rectification coefficient.

[0064] Please see also Figure 3 In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: dividing non-text data into image data, voice data, and public evaluation text; identifying the proportion of actual participants to expected participants in the image data using image detection technology; calculating the proportion of positive emotions in the voice data of the organizational life meeting using voice sentiment analysis technology; calculating the proportion of positive evaluations in the public evaluation text using text sentiment analysis technology; and calculating the non-text quantification value based on the proportion of actual participants to expected participants, the proportion of positive emotions in the voice data of the organizational life meeting, and the proportion of positive evaluations in the public evaluation text.

[0065] Specifically, the non-textual quantization value is calculated in the following way:

[0066] N = δ·Det(I activity O participant ) + ε·Sent(A meeting , S positive ) + ζ·Sent(C mass , S satisfy );

[0067] Where N is the non-text quantization value; Det(I activity O participant Sent(A) represents the proportion of actual participants to expected participants in an event photo identified by a large-scale image detection model. meeting , S positive Sent(C) represents the percentage of positive sentiment in organizational life meeting speech calculated by a large-scale speech sentiment analysis model. mass , S satisfy ) represents the percentage of positive evaluations in the public opinion text calculated by the large-scale text sentiment analysis model; δ represents the organizational life standardization coefficient; ε represents the ideological advancement coefficient; and ζ represents the public approval coefficient.

[0068] Specifically, for textual data such as work reports and ideological reports, the policy fit feature Fp, work effectiveness feature Fw, and problem rectification feature Fr are extracted using a large-scale text analysis model, and the text quantification value T is calculated using the following formula:

[0069] T = α·Sim(T text , T rule ) + β·TF-IDF(T text , K effect ) + γ·NER(T text E rectify );

[0070] Among them, Sim(T) text , T rule Sim(T) represents the semantic similarity between the report text and the core clauses of the regulations calculated by the large model. The semantic similarity is calculated using cosine similarity. Cosine similarity, also known as cosine similarity, assesses the similarity between two vectors by calculating the cosine of the angle between them. By creating two vectors for each text based on its keywords and calculating the cosine of these two vectors, the statistical similarity between the two texts can be determined. text ,T rule The value range of ) is [0,1]. For example, if there is content about rural revitalization work in a work report or a report on one's thoughts, the large-scale text analysis model will judge the matching degree between this part of the content and the rural revitalization clauses, and output the semantic similarity value.

[0071] TF-IDF(T text , K effect The TF-IDF weighted value represents the keywords related to work achievements in the text. TF-IDF (term frequency–inverse document frequency) is a commonly used weighting technique for information retrieval and data mining. TF stands for Term Frequency, and IDF stands for Inverse Document Frequency. In this embodiment, the keywords related to work achievements in the text include phrases such as "completed 3 community service projects" or "resolved 5 public requests." text , K effect The value range of TF-IDF(T) is [0,1]. In this embodiment, TF-IDF(T) text , K effect The numerical value reflects the degree of quantification of work performance.

[0072] NER(T text E rectifyThe value represents the quantified value of the problem rectification entity extracted by Named Entity Recognition (NER) in the large model. In this embodiment, Named Entity Recognition (NER) is a technique in natural language processing for identifying entities with specific meanings, such as names of people, places, and organizations, in text. NER technology can provide basic support for applications such as knowledge base question answering and machine translation. Through NER technology, quantified values ​​can be obtained for problem rectification entities, such as a 100% rectification completion rate or 0 non-compliant items, thus reflecting the effectiveness of problem resolution. In this embodiment, NER(T) represents the quantified value of the problem rectification entity extracted by Named Entity Recognition (NER) in the large model. text E rectify The value range of ) is [0,1].

[0073] α, β, and γ are Sim(T) text , T rule ), TF-IDF(T text , K effect ) and NER(T text E rectify The coefficient values ​​are α, β, and γ. In this embodiment, α, β, and γ satisfy the following relationship: α + β + γ = 1. The coefficient values ​​are determined by regulations prioritizing policy compliance > work performance > problem rectification. For example, in one embodiment, α = 0.4, β = 0.35, and γ = 0.25.

[0074] As needed, when calculating the text quantification value T, structured data accumulated from platform operations will also be analyzed simultaneously. This includes quantitative data such as learning duration, number of submitted reports, volunteer service duration, and number of meetings attended. This data, along with semantic similarity, TF-IDF weighted values ​​of work performance keywords, and the quantification value of problem rectification entities, serves as a supplementary reference for calculating the text quantification value, enhancing the comprehensiveness and accuracy of the evaluation data.

[0075] Specifically, for non-text data such as event photos and online meeting audio, a large-scale multimodal model is used to extract event participation features Fa, sentiment features Fe, and public satisfaction features Fm, and calculate the non-text quantification value N:

[0076] N = δ·Det(I activity O participant ) + ε·Sent(A meeting , S positive ) + ζ·Sent(C mass , S satisfy );

[0077] Among them, Det(I activity O participantThe det(I) represents the proportion of actual participants to the expected participants in an event photo identified by a large-scale image detection model. activity O participant The value range of Det(I) is [0,1]. Depending on the needs, Det(I) can be used... activity O participant It is necessary to exclude fabricated data from staged events.

[0078] Sent(A meeting , S positive The term "Sentiment Analysis" represents the proportion of positive emotions in the speech of an organizational life meeting, calculated by a large-scale speech sentiment analysis model. In this embodiment, sentiment analysis is a natural language processing method used to identify, extract, and quantify the emotional attitudes contained in text. The goal of sentiment analysis is to determine whether people's expressions in text are positive, negative, or neutral, and even to further distinguish more nuanced emotions such as anger, joy, sadness, and sarcasm. In this embodiment, the proportion of positive emotions in the speech of an organizational life meeting is calculated using a sentiment analysis model, such as whether someone proactively takes on a task or agrees with a decision. meeting , S positive The value range of Sent(A) is [0,1]. meeting ,S positive The numerical value reflects the dynamics of thought.

[0079] Sent(C mass , S satisfy Sent(C) represents the percentage of positive reviews in public opinion texts calculated by a large-scale text sentiment analysis model. mass , S satisfy The value range of Sent(C) is [0,1]. mass , S satisfy The numerical value represents the quantitative value of positive feedback from the public, such as "the problem was solved very promptly".

[0080] δ, ε, and ζ are respectively Det(I activity O participant ), Sent(A meeting , S positive ) and Sent(C mass ,S satisfy The coefficient values ​​are given by δ, ε, and ζ. In this embodiment, δ, ε, and ζ satisfy the following relationship: δ + ε + ζ = 1. The coefficient values ​​are determined with reference to the evaluation priority of organizational life standardization, ideological advancement, and public approval. In one embodiment, the evaluation priority is: organizational life standardization ≥ ideological advancement ≥ public approval. For example, δ = 0.35, ε = 0.3, and ζ = 0.35.

[0081] Please see also Figure 4 In some embodiments, the process of dynamically adjusting the weight values ​​of the policy interpretation intelligence, the data collection intelligence, and the indicator analysis intelligence according to different evaluation scenarios includes the following steps: setting the initial weights of the policy interpretation intelligence, the data collection intelligence, and the indicator analysis intelligence; in one evaluation scenario, calculating the weight of the policy interpretation intelligence based on the policy fit characteristics, the initial weight of the policy interpretation intelligence, and a first weight adjustment coefficient; calculating the weight of the data collection intelligence based on the activity participation characteristics, the initial weight of the data collection intelligence, and a second weight adjustment coefficient; and based on the... The weight of the indicator analysis agent is calculated based on the work performance characteristics, the initial weight of the indicator analysis agent, and the third weight adjustment coefficient. In another evaluation scenario, the weight of the policy interpretation agent is calculated based on the policy fit characteristics, the initial weight of the policy interpretation agent, and the first weight adjustment coefficient. The weight of the data collection agent is calculated based on the public satisfaction characteristics, the initial weight of the data collection agent, and the fourth weight adjustment coefficient. The weight of the indicator analysis agent is calculated based on the performance and dedication characteristics, the initial weight of the indicator analysis agent, and the fifth weight adjustment coefficient. Among these, the performance and dedication characteristics are extracted from volunteer activity records by the large model.

[0082] In this embodiment, based on the quantitative characteristics of multimodality, a dynamic weight calculation logic for multiple agents is designed, enabling the weights to adaptively adjust according to the evaluation scenario and regulatory priority, thereby achieving a balance between policy compliance and implementation feasibility. In this embodiment, the evaluation scenarios include compliance review or performance evaluation, etc.

[0083] Specifically, when calculating the fortress index, the corresponding association weights are calculated in the following way.

[0084] The initial weight of the policy interpretation agent is set to Wp0. The initial weight of the data collection agent is set to Wd0. The initial weight of the indicator analysis agent is set to Wa0. The initial weights of each agent are anchored to regulations. In one embodiment, the priority of the initial weights is set as follows: policy compliance ≥ data authenticity ≥ indicator analytical capability. For example, in one embodiment, Wp0 = 0.4, Wd0 = 0.3, and Wa0 = 0.3.

[0085] Let the weight of the policy interpretation agent be Wp, the weight of the data collection agent be Wd, and the weight of the indicator analysis agent be Wa. The specific calculation formulas for Wp, Wd, and Wa are as follows: Wp = Wp0·(1 + λ·Fp); Wd = Wd0·(1 + μ·Fa); Wa= Wa0·(1 + ν·Fw).

[0086] Wherein, λ is the first weight adjustment coefficient; μ is the second weight adjustment coefficient; and ν is the third weight adjustment coefficient. The first weight adjustment coefficient λ represents the degree of influence of the policy fit characteristic Fp on the weight Wp of the policy interpretation agent. The second weight adjustment coefficient μ represents the degree of influence of the activity participation characteristic Fa on the weight Wd of the data collection agent. The third weight adjustment coefficient ν represents the degree of influence of the work performance characteristic Fw on the weight Wa of the indicator analysis agent.

[0087] Specifically, when calculating the Vanguard Index, the corresponding association weights are calculated in the following way.

[0088] For another evaluation scenario, the influence of public evaluation and performance of duties and dedication is strengthened, and the weight calculation logic is adjusted: Wp = Wp0·(1 + λ·Fp); Wd = Wd0·(1 + μ'·Fm); Wa = Wa0·(1 + ν'·Fv).

[0089] Where μ' is the fourth weight adjustment coefficient and ν' is the fifth weight adjustment coefficient. The fourth weight adjustment coefficient μ' represents the degree of influence of the public satisfaction feature Fm on the weight Wd of the data collection agent. The fifth weight adjustment coefficient ν' represents the degree of influence of the performance and dedication feature Fv on the weight Wa of the indicator analysis agent.

[0090] In this embodiment, Fm represents the public satisfaction feature extracted from the public review text by the large model analysis, with a value range of [0,1]. The value of the public satisfaction feature Fm is a quantitative value of the public's comprehensive evaluation of "service response speed" and "problem-solving quality".

[0091] Fv represents the performance and dedication features extracted from volunteer activity records in a large-scale model analysis, with a value range of [0,1]. Volunteer activity records include activity photos, service duration text, etc. The value of the performance and dedication feature Fv is a fusion value of the annual volunteer service duration percentage, the service recipient satisfaction rate, etc.

[0092] Please see also Figure 5 In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: in one evaluation scenario, under the premise of meeting the discipline and compliance dimension, the first weight adjustment coefficient is automatically increased to the sixth weight adjustment coefficient to increase the weight of the policy interpretation agent; in another evaluation scenario, when the large model detects a tendency for learning and education to be substandard, the first weight adjustment coefficient is automatically increased to the seventh weight adjustment coefficient to increase the weight of the policy interpretation agent.

[0093] Specifically, when calculating the fortress index, if the large model detects a risk in the discipline compliance dimension, the model will automatically increase the first weight adjustment coefficient λ to 0.8 and simultaneously increase the Wp weight to ensure that policy compliance is given priority in the judgment.

[0094] Specifically, when calculating the Pioneer Index, if the large model detects a tendency to fail to meet learning and education standards, such as failing to complete learning tasks for three consecutive months or having learning time below the average level, the sub-weight of the obligation clause analysis module in the policy interpretation agent is automatically increased (the sub-weight ratio is increased from 20% to 40%) to strengthen compliance warnings.

[0095] In some embodiments, the large-scale model multi-agent system further includes an expert interactive agent. This expert interactive agent generates correction coefficients to adjust the information evaluation results. The expert agent can be used in specific scenarios as needed.

[0096] Please see also Figure 6 In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following step: obtaining expert opinions through the expert interactive agent;

[0097] The suitability of expert opinions to the scenario to be evaluated is calculated using semantic analysis techniques; the weighting coefficient of expert opinions is determined based on the experts' years of experience and the richness of their evaluation experience; the correction coefficient is generated based on the suitability of expert opinions to the scenario to be evaluated and the weighting coefficient of expert opinions; and the corrected information evaluation result is determined based on the correction coefficient and the information evaluation result.

[0098] Specifically, the correction factor is calculated as follows: C exp = 1 + θ·Reason(E opinion , S scene ).

[0099] Among them, C exp This is the correction factor; Reason(E) opinion , S scene ) represents the fit between expert opinions and specific scenarios calculated by the large-scale semantic analysis model; θ represents the weight coefficient of expert opinions.

[0100] The revised information assessment result is calculated in the following way: R final = R pre ·C exp .

[0101] Among them, R final The revised information evaluation results; R pre The information assessment results are as shown before the revision.

[0102] In this embodiment, experts can conduct precise analysis and quantitative correction of the information evaluation results. Experts can combine specific scenarios to adjust R... pre Make corrections and generate correction coefficient C. exp The formula is as follows:

[0103] C exp = 1 + θ·Reason(E opinion , S scene );

[0104] Reason(E) opinion , S scene Reason(E) represents the expert opinions and their fit to specific scenarios calculated by the semantic analysis model of a large model. opinion , S scene The value range of ) is [-0.2, 0.2]. If an expert suggests adjusting the participation rate based on online participation, the model calculates the degree of fit between this opinion and the management scenario: Reason(E). opinion , S scene ), and generate the correction coefficient C accordingly. exp .

[0105] θ represents the expert opinion weighting coefficient. The value of the expert opinion weighting coefficient θ ranges from [0.5, 1]. The expert opinion weighting coefficient θ is determined by the superior authority based on factors such as the expert's years of service and experience level. For example, for experts with more than 10 years of experience, the expert opinion weighting coefficient θ is θ = 0.9.

[0106] Final evaluation result R final = R pre ·C exp In this embodiment, R final The threshold range set by the regulations must be met. For example, when calculating the Pioneer Index, a Pioneer Index ≥ 80 is considered excellent, and a Pioneer Index < 60 is considered to need improvement.

[0107] Please see also Figure 7 In some embodiments, the information management method based on a multi-agent collaborative decision-making mechanism further includes the following steps: setting a reward signal, wherein the reward signal is derived by combining the degree of consistency between the corrected information evaluation result and the subsequent actual performance result after a preset time and the proportion of the number of expert corrections to the total number of evaluations; and iterating on any one or more of the first weight adjustment coefficient, the second weight adjustment coefficient, the third weight adjustment coefficient, the fourth weight adjustment coefficient, and the fifth weight adjustment coefficient according to the reward signal.

[0108] In this embodiment, the Temporal Differential Learning (TD-Learning) algorithm is introduced, with the accuracy of the evaluation result as the objective function, and the weight adjustment coefficients (λ, μ, ν, μ', ν') are iteratively optimized based on expert correction feedback and subsequent tracking data.

[0109] Specifically, the reward function is designed first. The reward signal (Reward) is set as the degree of agreement between the evaluation result and the subsequent actual performance. The formula for calculating the reward signal (Reward) is as follows:

[0110] Reward = η·Match(R final , P subseq ) + (1-η)·(1 - Revise(E times ));

[0111] Match(R) final , P subseq R represents the final evaluation result of the large model analysis. final The degree of consistency with actual performance 3 months later. Actual performance 3 months later includes awards received or completion of corrective actions, etc. Match(R) final , P subseq The value range of ) is [0,1].

[0112] Revise(E times The number of revisions represents the proportion of expert corrections to the total number of evaluations. Fewer revisions indicate a more accurate pre-analysis of the model. Revise(E) times The value range of ) is [0,1].

[0113] η represents the result consistency reward weighting coefficient. The value range of the result consistency reward weighting coefficient η is [0.6, 0.8]. The result consistency reward weighting coefficient η is used to prioritize ensuring the actual validity of the evaluation results.

[0114] Then, the weights are iteratively updated.

[0115] Taking the "policy fit adjustment coefficient λ" as an example, the iterative formula is as follows:

[0116] λ(t+1) = λ(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), λ(t)) - Q^(s(t), λ(t))];

[0117] λ(t) is the first weight adjustment coefficient for policy fit before iteration; λ(t+1) is the first weight adjustment coefficient for policy fit after iteration. σ represents the learning rate. The learning rate σ ranges from [0.01, 0.1]. The learning rate σ is used to control the iteration step size to avoid excessive weight fluctuations. τ represents the discount factor. The discount factor τ ranges from [0.7, 0.9]. The discount factor τ is used to balance current rewards with future long-term rewards. Q^(s(t), λ(t)) represents the action value function corresponding to the coefficient λ(t) in state s(t), reflecting the contribution of this coefficient to the reward. The state s(t) includes the characteristics of the current evaluation scenario.

[0118] Other adjustment coefficients (μ, ν, μ', ν') are iterated using the same logic to ensure that the weight model is continuously optimized as the scenario changes.

[0119] For example, the iterative formula for the activity participation adjustment coefficient μ is as follows:

[0120] μ(t+1) = μ(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), μ(t)) - Q^(s(t), μ(t))].

[0121] For the performance adjustment coefficient ν, the iterative formula is as follows:

[0122] ν(t+1) = ν(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), ν(t)) - Q^(s(t), ν(t))].

[0123] For the adjustment coefficient ν of public satisfaction, the iterative formula is as follows:

[0124] μ'(t+1) = μ'(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), μ'(t)) - Q^(s(t), μ'(t))].

[0125] For the performance contribution adjustment coefficient ν, the iterative formula is as follows:

[0126] ν'(t+1) = ν'(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), ν'(t)) - Q^(s(t), ν'(t))].

[0127] In this embodiment, based on the reward signal, any one or more of the first weight adjustment coefficient, the second weight adjustment coefficient, the third weight adjustment coefficient, the fourth weight adjustment coefficient, and the fifth weight adjustment coefficient are iterated. This method achieves human-machine collaborative decision-making and enables the process to achieve an efficient closed loop. In existing information management systems, the integration of "large model + human" remains at a shallow stage of model output and independent human decision-making. Pre-analysis lacks quantitative support, feedback lacks an iterative mechanism, resulting in low decision-making efficiency and recurring biases. This application achieves efficient integration and continuous optimization of human-machine collaboration through quantitative collaborative processes and reinforcement learning iteration. In one embodiment, the information management system also includes a full-link traceability and responsibility delineation submodule, used to record key information at each stage, forming a traceable closed-loop link. By analyzing the entire link record through model analysis, the root cause of decision-making biases (such as multimodal feature extraction errors or human correction mistakes) can be traced, solving the problem of ambiguous human-machine responsibility delineation in existing technologies and improving the credibility of technology applications. In terms of decision-making efficiency, large-scale model joint pre-analysis can automatically integrate multimodal data, verify policy compliance, and generate structured pre-assessment results with accompanying data and compliance reports. Experts do not need to sift through information from scratch; they only need to make corrections for specific scenarios, significantly shortening the time for a single decision and greatly improving efficiency. In terms of decision-making accuracy, the introduction of expert correction coefficients and reward signals transforms expert experience into calculable parameters, such as scenario suitability. Then, through algorithmic iteration to optimize the weight adjustment coefficients, the expert correction rate in model pre-analysis is significantly reduced, minimizing the recurrence of similar decision-making biases.

[0128] Please see Figure 8 Another embodiment of this application provides an information management system 100 based on a multi-agent collaborative decision-making mechanism. The information management system 100 includes a large-model multi-agent module 110, a dynamic weight decision-making module 120, and a multi-agent collaborative decision-making module 130.

[0129] The large-scale model multi-agent module 110 includes a policy interpretation agent 111, a data collection agent 112, an indicator analysis agent 113, and an expert interaction agent 114. The policy interpretation agent 111 outputs compliance pre-assessment results. The data collection agent 112 outputs data authenticity pre-assessment results. The indicator analysis agent outputs advancement pre-assessment results.

[0130] The data acquisition agent 112 is used to collect multimodal data input by the user and to quantify the text and non-text data within the multimodal data. During the quantification of text data, policy alignment features, work effectiveness features, and problem rectification features are extracted from the text data, and a text quantification value is calculated based on these features. During the quantification of non-text data, activity participation features, sentiment characteristics, and public satisfaction features are extracted from the non-text data, and a non-text quantification value is calculated based on these features.

[0131] The data acquisition agent 112 outputs a pre-assessment result of data authenticity based on the data consistency and data integrity of the multimodal data. The policy interpretation agent 111 outputs a pre-assessment result of compliance based on the policy fit characteristics and the number of matching compliance clauses. The indicator analysis agent 113 outputs a pre-assessment result of advancement based on the analysis indicators of the assessment scenario.

[0132] The dynamic weight decision module 120 is used to dynamically adjust the weight values ​​of the policy interpretation agent 111, the data collection agent 112, and the indicator analysis agent 113 according to different evaluation scenarios.

[0133] The multi-agent collaborative decision-making module 130 is used to generate information evaluation results based on the weight values ​​of the policy interpretation agent 111, the data collection agent 112, the indicator analysis agent 113, the compliance pre-evaluation results, the data authenticity pre-evaluation results, and the advancement pre-evaluation results.

[0134] In some embodiments, the multi-agent collaborative decision-making module 130 includes a large-scale model joint pre-analysis submodule 131, an expert correction and quantification submodule 132, a decision result output and application submodule 133, and a full-link traceability and responsibility delineation submodule 134. The large-scale model joint pre-analysis submodule 131 generates information evaluation results based on the weight values ​​of the policy interpretation agent 111, the data collection agent 112, the indicator analysis agent 113, and the compliance pre-assessment results, data authenticity pre-assessment results, and advancement pre-assessment results. The expert correction and quantification submodule 132 generates correction coefficients to correct the information evaluation results. The decision result output and application submodule 133 generates a standardized evaluation report adapted to multiple terminals from the corrected information evaluation results. The full-link traceability and responsibility delineation submodule 134 records key information at each stage to form a traceable link.

[0135] The information management system 100 provided in this embodiment deeply integrates large-scale model multi-agent technology and multimodal processing technology, constructing a four-layer progressive architecture of "multimodal data layer - specialized agent layer - dynamic decision-making layer - scenario-based application". Each layer achieves bidirectional interaction through standardized data interfaces, forming a full-process intelligent closed loop of data acquisition - quantitative processing - collaborative decision-making - result output - model iteration.

[0136] The multimodal data layer is responsible for accessing raw data such as text reports, images of events, and audio recordings of organizational activities, providing a comprehensive data source for subsequent processing. The specialized intelligent agent layer deploys four types of intelligent agents: policy interpretation, data collection, indicator analysis, and expert interaction, each undertaking core functions such as regulation interpretation, multimodal data quantification, dual-index calculation, and human-machine collaborative communication. The dynamic decision-making layer relies on a dual-index-driven dynamic weight adaptive model to adjust the weights of the intelligent agents, combining large-scale model joint pre-analysis and expert fine-tuning to generate the final evaluation results, and iteratively optimizing the weight coefficients through a reinforcement learning engine. The scenario-based application layer outputs "Fortress Index" and "Pioneer Index" evaluation reports, adapting to multi-level management needs and ensuring the compliance, accuracy, and practical adaptability of the technical solution.

[0137] The overall workflow of the information management system 100 is as follows: Figure 10 As shown in the diagram, the data acquisition agent 112 is used for multimodal data acquisition and quantification. As the core functional carrier for system data input and standardized preprocessing, the data acquisition agent 112 focuses on the data integration needs of the entire management process. Through a three-level processing chain of "multi-source acquisition - cleaning and verification - feature quantification," it solves the problems of scattered grassroots data, low utilization rate of unstructured data, and difficulty in quality control, providing standardized data support for subsequent collaborative decision-making.

[0138] Please see also Figure 9The data acquisition intelligent agent 112 includes a multi-terminal data access submodule 1121, a data quality control submodule 1122, and a multimodal feature quantification submodule 1123. The multi-terminal data access submodule 1121 supports collaborative data acquisition from three terminals: computer, mobile phone, and smart screen, covering three core data types: text, image, and voice. For example, text data can be uploaded via a document upload interface on the computer or edited and submitted online on a mobile phone. The text data format is compatible with DOC / PDF / TXT formats. Uploaded text data can be automatically associated with metadata such as submission time, as needed. Image data, such as photos of events or videos of volunteer practices, can be uploaded via mobile phone photography or imported via QR code scanning on a smart screen. After image data is uploaded, the system automatically matches the event theme and participant information. Voice data, such as recordings of organizational meetings or lecture audio, can be recorded in real-time via a terminal microphone or imported from a local file. As needed, the system will synchronously mark key audio nodes, such as policy interpretation paragraphs and discussion / resolution stages, forming a closed loop of business data throughout the entire process. The data quality control submodule 1122 uses large-scale modeling technology to screen data compliance and authenticity. For example, text data is identified for plagiarism and missing elements (such as reports without dates) through semantic analysis. Content with a duplication rate exceeding 80% will be automatically flagged and a supplementary reminder will be triggered. Image data is filtered to eliminate staged images. For example, if the background does not match the event scene, the corresponding image data will be automatically excluded. As needed, the data quality control submodule 1122 can also determine the number of attendees through image detection and compare it with attendance data for consistency. When voice data is transcribed into text data, the data quality control submodule 1122 verifies whether it contains core knowledge points, key meeting resolutions, etc., ensuring that the data meets the requirements for authenticity and completeness. The multimodal feature quantification submodule 1123 converts data into standardized, computable parameters according to data type. For text data, the multimodal feature quantification submodule 1123 extracts core features based on policy relevance, work effectiveness, and problem rectification. This is achieved using the following formula:

[0139] T = α·Sim(T text , T rule ) + β·TF-IDF(T text , K effect ) + γ·NER(T text E rectify );

[0140] The multimodal feature quantization submodule 1123 can calculate the quantization value of text data.

[0141] For non-text data, the multimodal feature quantization submodule 1123 extracts core features based on activity participation rate, sentiment tendency, and public satisfaction. For example, activity participation rate includes the percentage of actual participants in the photos. Sentiment tendency includes the level of participation in meeting discussions. Public satisfaction includes the positive review rate of public comments. This is achieved through the following formula:

[0142] N = δ·Det(I activity O participant ) + ε·Sent(A meeting , S positive ) + ζ·Sent(C mass , S satisfy );

[0143] The multimodal feature quantization submodule 1123 can convert image and voice information into standardized parameters in the [0,1] interval, providing a data foundation for evaluation.

[0144] In this embodiment, the dynamic weight decision-making module 120 is constructed based on the differences in evaluation scenarios and the priority requirements of regulations. By constructing a dynamic weight calculation and iterative optimization mechanism, the agent's weights are adaptively adjusted according to the evaluation scenario. At the same time, the weight model is continuously optimized by relying on reinforcement learning algorithms, solving the problems of traditional fixed weight models being unable to adapt to multiple scenarios and having low decision-making accuracy, thus achieving a balance between policy compliance and feasibility of implementation at the grassroots level.

[0145] In this embodiment, the dynamic weight decision module 120 includes a fortress index weight calculation submodule 121, a vanguard index weight calculation submodule 122, and a reinforcement learning weight iteration submodule 123.

[0146] The fortress index weight calculation submodule 121 is used to set the weights Wp, Wd, and Wa of the policy interpretation agent, data collection agent, and indicator analysis agent. In this embodiment, the fortress index weight calculation submodule 121 is based on the baseline weight anchored by regulations (e.g., Wp0=0.4, prioritizing policy compliance), and adjusts the weights in combination with features such as policy fit, activity participation, and work effectiveness (e.g., task completion quality). For example, when evaluating the effectiveness of rural revitalization, increasing the weight of the policy fit feature increases Wp to focus on verifying whether the activities meet the requirements.

[0147] The vanguard index weight calculation submodule 122 is used to optimize the weight logic for another evaluation scenario, strengthening the influence of "public evaluation" and "performance of duties and contributions." Specifically, the weight of the data collection agent 112 is correlated with public satisfaction characteristics, such as the public's positive feedback rate on services, and is dynamically adjusted using the formula Wd = Wd0·(1 + μ'·Fm). The weight of the indicator analysis agent 113 is correlated with performance of duties and contributions characteristics, such as the percentage of annual volunteer service hours and the positive feedback rate of service recipients, and is dynamically adjusted using the formula Wa = Wa0·(1 + ν'·Fv). When substandard learning and education are detected, such as failure to complete learning tasks for three consecutive months, the weight of the obligation clause analysis in the policy interpretation agent 111 is automatically increased to strengthen compliance warnings and ensure that the evaluation takes into account both public approval and policy requirements.

[0148] The reinforcement learning weight iteration submodule 123 introduces a temporal difference learning (TD-Learning) algorithm to optimize the weight adjustment coefficients with the accuracy of the evaluation results as the objective function. Specifically, the reinforcement learning weight iteration submodule 123 designs a reward function:

[0149] Reward = η·Match(R final , P subseq ) + (1-η)·(1 - Revise(E times The model comprehensively considers the consistency between the evaluation results and subsequent actual performance, as well as the correlation with the expert correction rate. Fewer expert corrections indicate a more accurate pre-analysis. The reinforcement learning weight iteration submodule 123 dynamically updates the coefficients using an iterative formula.

[0150] λ(t+1) = λ(t) +σ·[Reward(t+1) +τ·Q^(s(t+1), λ(t)) - Q^(s(t), λ(t))];

[0151] At the same time, by controlling the iteration step size σ and the future reward weight τ, the weight model can continuously adapt to changes in the scenario.

[0152] The multi-agent collaborative decision-making module 130 is used to construct a closed-loop mechanism covering the entire process of "large model pre-analysis - expert in-depth judgment - result output - model iteration". In this embodiment, the multi-agent collaborative decision-making module 130 integrates intelligent document generation and full-link traceability functions, realizing a deep integration of large model technology and expert experience. In other words, the multi-agent collaborative decision-making module 130 can solve the problems of shallow human-machine collaboration, lack of feedback iteration, and ambiguous responsibility definition in traditional human-machine collaboration, thereby improving the efficiency and credibility of decision-making.

[0153] In this embodiment, the multi-agent collaborative decision-making module 130 includes a large model joint pre-analysis sub-module 131, an expert correction quantification sub-module 132, a decision result output and application sub-module 133, and a full-link traceability and responsibility delineation sub-module 134.

[0154] The large model joint pre-analysis submodule 131 integrates the output data of the three types of agents to generate a dual-exponential pre-evaluation result, which is expressed by formula R. pre The preliminary score is calculated as Wp·Rp + Wd·Rd + Wa·Ra. Specifically, policy interpretation agent 111 provides a compliance preliminary score (Rp, based on policy fit and the number of clause matches), data collection agent 112 provides a data authenticity preliminary score (Rd, based on multimodal data consistency and completeness), and indicator analysis agent 113 provides an advancement preliminary score (Ra, based on dual-index core indicators). The total preliminary score is calculated by combining dynamic weights (Wp, Wd, Wa) to generate the information evaluation result. The information evaluation result includes detailed data support, such as screenshots of learning time and activity photos, providing sufficient support for expert judgment.

[0155] The expert correction and quantification submodule 132 supports experts in correcting information evaluation results for specific scenarios, and then generates corrected information evaluation results through a formula. In this embodiment, the expert correction and quantification submodule 132 combines the adaptability of expert opinions to specific scenarios (Reason, such as the adaptability of "adjusting participation rate based on online participation" to the management scenario), and the expert's experience level (θ, such as θ=0.9 for experts with more than 10 years of experience), and uses formula C to correct the results. exp = 1 + θ·Reason(E opinion ,S scene Calculate the correction coefficients; finally, obtain the corrected information evaluation result R. final = R pre ·C exp The revised information assessment results must meet the threshold range set by the regulations (e.g., a pioneer index ≥80 is excellent, <60 is in need of improvement) to ensure that the assessment results are accurate and compliant in special scenarios.

[0156] The decision result output and application submodule 133 is used to generate standardized evaluation reports adapted to multiple terminals. This submodule 133 can synchronously support the entire management process, such as exporting PDF / Excel reports on the computer and automatically generating reports. On the smart screen, the distribution of the fortress index and vanguard index is displayed using visual charts (score radar chart, ranking bar chart). On the mobile device, personal evaluation results and improvement suggestions are pushed, such as increasing learning time. The decision result output and application submodule 133 can also simultaneously trigger intelligent copywriting functionality, automatically generating work summaries and activity briefings, covering dual-index analysis and typical case studies, reducing the workload of copywriting.

[0157] The end-to-end traceability and responsibility delineation submodule 134 records key information from each stage, including data collection, quantification, weight calculation, pre-analysis, and expert correction, forming a traceable chain. Each evaluation result is associated with the multimodal raw data ID, agent analysis log, and expert correction record (including correction time, reason, and content). When evaluation deviations occur, the problematic stage can be traced back to pinpoint (e.g., data quantification error, unreasonable weight coefficients, or logical deviations in expert correction), clearly defining the responsibility of the technical algorithm and human judgment, and avoiding the phenomenon of human-machine mutual blame.

[0158] As needed, the information management system 100 based on a multi-agent collaborative decision-making mechanism may further include a multi-terminal adaptation and data value-added application module. This module addresses the collaborative needs of three terminals—computer, mobile phone, and smart screen—achieving deep functional and scenario adaptation. Furthermore, it can expand data value-added applications, resolving issues such as significant performance differences among grassroots terminals and low data value utilization, covering the entire management process, and driving a shift in management from experience-driven to data-driven approaches.

[0159] In this embodiment, the multi-terminal adaptation and data value-added application module includes a multi-terminal function adaptation sub-module, a basic terminal lightweight adaptation sub-module, and a data value-added application sub-module.

[0160] The multi-terminal function adaptation submodule is designed to provide differentiated functions for different terminal characteristics, meeting the needs of users at multiple levels. The PC version (Big Data Platform) offers complex operation functions such as report generation, intelligent document assistant, and multi-agent management, supporting batch data processing and visualization analysis (such as monthly task completion rate charts and user profile tag displays). The mobile version's "Cultivation Assistant" mini-program focuses on lightweight operation, supporting the submission of ideological reports, registration for volunteer activities, and querying personal indices. It features a simplified interface, larger fonts, and voice navigation. The smart screen version is deployed in the activity room, showcasing the branch's style and displaying activity registration QR codes. Users can scan the code to participate in activities, and the screen simultaneously plays videos, policy interpretations, and other learning resources, enriching offline scenarios.

[0161] The lightweight adaptation submodule for basic terminals is designed for low-configuration terminals and simple devices. This submodule optimizes the technical architecture to ensure smooth operation on ordinary computers and smartphones. Simultaneously, it supports offline data caching; for example, photos from offline volunteer activities can be temporarily stored on the phone and automatically synchronized to the system after connecting to the internet.

[0162] The data value-added application submodule constructs profiles and management models based on evaluation data, supporting precise and personalized services. For example, this submodule can generate learning preference tags to push personalized learning resources. Simultaneously, it can identify advanced organizations, extract their activity organization models and policy implementation methods, and generate experience promotion reports. This submodule can also provide the big data command center with visualized data such as monthly task completion rates and the progress of training targets, dynamically showcasing work effectiveness and providing data support for decision-making.

[0163] Existing management systems are mostly limited to structured data processing and fail to consider the performance differences of grassroots terminals, resulting in unrealized data value and the inability to utilize advanced technologies. The information management system 100 provided in this application achieves a dual improvement in data value and grassroots usability through a multimodal quantification mechanism and a lightweight adaptive design. This mechanism can significantly improve the completeness of grassroots data collection and the system usage rate for special groups, expanding the coverage of technology applications.

[0164] In terms of data value, the information management system 100 provided in this application breaks through the limitations of using only structured data. By using large-scale model multimodal technology, it can mine implicit values ​​such as the depth of policy understanding in ideological reports, the authenticity of participation in activity photos, and the emotional tendencies in public evaluations. This greatly expands the scope of data utilization and provides more comprehensive data support for dual-index evaluation and dynamic decision-making, such as the fortress index and the vanguard index.

[0165] At the grassroots adaptation level, the information management system 100 provided in this application optimizes multimodal models for scenarios such as low-configuration terminals and simple devices by employing lightweight optimization techniques, such as knowledge distillation to compress large model parameters, thereby ensuring that complex algorithms can run smoothly on ordinary computers and smartphones. Simultaneously, the information management system 100 also supports offline data caching, such as temporarily storing offline activity photos and then uploading them in batches, thus addressing the digital divide problem in existing information management software where it is unusable at the grassroots level and data cannot be transmitted.

[0166] Please see also Figure 11Another embodiment of this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the above embodiments.

[0167] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An information management method based on a multi-agent collaborative decision-making mechanism, characterized in that, Includes the following steps: Construct a large-scale model multi-agent system, which includes a policy interpretation agent, a data collection agent, and an indicator analysis agent; The data acquisition agent is used to collect multimodal data input by the user, including text data and non-text data; The text data is quantified by extracting policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics, and calculating the text quantification value based on the policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics; the non-text data is also quantified by extracting activity participation characteristics, sentiment characteristics, and public satisfaction characteristics, and calculating the non-text quantification value based on the activity participation characteristics, sentiment characteristics, and public satisfaction characteristics. The data collection agent outputs a preliminary assessment result on the authenticity of the data; the policy interpretation agent outputs a preliminary assessment result on compliance; and the indicator analysis agent outputs a preliminary assessment result on advancement. The weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent are dynamically adjusted according to different evaluation scenarios. Information evaluation results are generated based on the weight values ​​of the policy interpretation intelligence agent, the data collection intelligence agent, the indicator analysis intelligence agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results.

2. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 1, characterized in that, It also includes the following steps: Compare the text data with the core clauses of the regulations to obtain the semantic similarity between the text data and the core clauses of the regulations; Extract keywords related to work performance from the text data, and calculate their TF-IDF weighted values ​​based on these keywords. The quantitative values ​​of the problem rectification entity in the text data are extracted using named entity recognition technology; The text quantification value is calculated based on the semantic similarity, the TF-IDF weighted value of the keywords of the work results, and the quantification value of the problem rectification entity.

3. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 1, characterized in that, It also includes the following steps: Non-text data is categorized into image data, audio data, and public comment texts. Image detection technology is used to identify the proportion of actual participants to the expected participants in image data. The proportion of positive emotions in the speech of organizational life meetings was calculated using speech sentiment analysis technology. The percentage of positive reviews in public opinion texts was calculated using text sentiment analysis technology. The non-textual quantitative value is calculated based on the proportion of actual participants to the expected participants, the proportion of positive emotions in the audio recordings of the organizational life meeting, and the proportion of positive evaluations in the public review texts.

4. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 1, characterized in that, The process of dynamically adjusting the weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent according to different evaluation scenarios includes the following steps: Set the initial weights for the policy interpretation agent, the data collection agent, and the indicator analysis agent; In one evaluation scenario, the weight of the policy interpretation agent is calculated based on the policy fit characteristics, the initial weight of the policy interpretation agent, and a first weight adjustment coefficient; the weight of the data collection agent is calculated based on the activity participation characteristics, the initial weight of the data collection agent, and a second weight adjustment coefficient; and the weight of the indicator analysis agent is calculated based on the work effectiveness characteristics, the initial weight of the indicator analysis agent, and a third weight adjustment coefficient. In another evaluation scenario, the weight of the policy interpretation agent is calculated based on the policy fit characteristic, the initial weight of the policy interpretation agent, and the first weight adjustment coefficient; the weight of the data collection agent is calculated based on the public satisfaction characteristic, the initial weight of the data collection agent, and the fourth weight adjustment coefficient; and the weight of the indicator analysis agent is calculated based on the performance and dedication characteristic, the initial weight of the indicator analysis agent, and the fifth weight adjustment coefficient. The performance and dedication characteristic is extracted from volunteer activity records by the large model.

5. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 4, characterized in that, It also includes the following steps: In one assessment scenario, provided that the discipline and compliance dimension must be met, the first weight adjustment coefficient is automatically increased to the sixth weight adjustment coefficient to enhance the weight of the policy interpretation agent. In another evaluation scenario, when a tendency for learning and education to fall short of standards is detected, the first weight adjustment coefficient is automatically increased to the seventh weight adjustment coefficient to enhance the weight of the policy interpretation agent.

6. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 4, characterized in that, The large-scale model multi-agent system also includes an expert interactive agent, which is used to generate correction coefficients to correct the information evaluation results.

7. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 6, characterized in that, It also includes the following steps: Obtain expert opinions through the aforementioned expert interactive intelligent agent; The suitability of expert opinions to the scenario to be evaluated is calculated based on semantic analysis techniques. The weighting coefficient of expert opinions is determined based on the experts' years of service and the richness of their evaluation experience. The correction coefficient is generated based on the fit between expert opinions and the scenario to be evaluated, and the weighting coefficient of expert opinions. The corrected information evaluation result is determined based on the correction coefficient and the information evaluation result.

8. The information management method based on a multi-agent collaborative decision-making mechanism according to claim 7, characterized in that, It also includes the following steps: A reward signal is set, which is derived by combining the degree of consistency between the corrected information evaluation result and the subsequent actual performance result after a preset time, as well as the proportion of the number of expert corrections to the total number of evaluations. Based on the reward signal, iterate over any one or more of the first weight adjustment coefficient, the second weight adjustment coefficient, the third weight adjustment coefficient, the fourth weight adjustment coefficient, and the fifth weight adjustment coefficient.

9. An information management system based on a multi-agent collaborative decision-making mechanism, characterized in that, include: The large-scale model multi-agent module includes a policy interpretation agent, a data collection agent, and an indicator analysis agent. The data acquisition agent is used to collect multimodal data input by users and to quantify the text and non-text data within the multimodal data. During the quantification of text data, policy fit characteristics, work effectiveness characteristics, and problem rectification characteristics are extracted, and a text quantification value is calculated based on these characteristics. During the quantification of non-text data, activity participation characteristics, sentiment characteristics, and public satisfaction characteristics are extracted, and a non-text quantification value is calculated based on these characteristics. The data acquisition agent outputs a pre-assessment result of data authenticity based on the data consistency and integrity of the multimodal data. The policy interpretation agent outputs a pre-assessment result of compliance based on the policy fit characteristics and the number of matching compliance clauses. The indicator analysis agent outputs a pre-assessment result of advancement based on the analytical indicators of the assessment scenario. The dynamic weight decision module is used to dynamically adjust the weight values ​​of the policy interpretation agent, the data collection agent, and the indicator analysis agent according to different evaluation scenarios. The multi-agent collaborative decision-making module is used to generate information evaluation results based on the weight values ​​of the policy interpretation agent, the data collection agent, the indicator analysis agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results.

10. The information management system based on a multi-agent collaborative decision-making mechanism according to claim 9, characterized in that, The large model multi-agent system also includes an expert interactive agent, which is used to generate correction coefficients to correct the information evaluation results; The multi-agent collaborative decision-making module includes: The large model joint pre-analysis submodule is used to generate information evaluation results based on the weight values ​​of the policy interpretation agent, the data collection agent, the indicator analysis agent, the compliance pre-assessment results, the data authenticity pre-assessment results, and the advancement pre-assessment results. The expert-corrected quantification submodule is used to generate correction coefficients to correct the information evaluation results; The decision result output and application submodule is used to generate a standardized evaluation report adapted to multiple terminals from the corrected information evaluation results. The end-to-end traceability and responsibility delineation submodule is used to record key information at each stage to form a traceable chain.

Images