Data matching method and system based on AI model and knowledge graph driving

Abstract

The invention relates to the crossing field of artificial intelligence and big data, and discloses a data matching method based on an AI model and knowledge graph driving, which comprises the following steps: intelligently capturing policy data, collecting enterprise data, and carrying out document processing and structured storage; constructing a policy knowledge graph and realizing enterprise data enhancement, including knowledge base initialization and AI auxiliary modeling; a mixed matching engine and a rule reasoning mechanism are integrated, dynamic and accurate matching of enterprises and policies is achieved, a causal discovery algorithm module is embedded in the mixed matching engine and used for recognizing the causal relationship between policy terms and enterprise behaviors, mismatching caused by false correlation is avoided, and the purposes of improving the intelligent level of policy matching and improving the policy matching efficiency are achieved. Personalized and high-precision policy recommendation support is provided for enterprises, meanwhile, scientificity and effectiveness of government resource allocation are improved, and good practical value and popularization prospects are achieved.

Description

Technical Field

[0001] This invention relates to the intersection of artificial intelligence and big data, and more specifically, to a data matching method and system driven by AI models and knowledge graphs. Background Technology

[0002] Against the backdrop of the current deep integration of policy promotion and enterprise development, the government has introduced a large number of industry support policies, special fund plans, and technology innovation guidance programs, aiming to provide precise support to enterprises of different types and stages. However, the sheer number, complexity, and frequent updates of these policies have led to difficulties for enterprises in obtaining information, interpreting policies, and determining their suitability during the application process. Simultaneously, the government also faces the challenge of achieving precise policy implementation and efficient use of funds.

[0003] In existing technologies, some systems attempt to assist policy recommendations through keyword matching, rule base filtering, and other methods, but these methods generally have the following limitations: The data sources are singular and the information is incomplete. It relies heavily on information proactively submitted by companies and lacks the ability to automatically obtain publicly available data from enterprises. The matching mechanism is crude and lacks semantic understanding. It often relies on simple text comparison methods, failing to gain a deep understanding of the implicit relationship between policy provisions and the actual capabilities of enterprises. Knowledge graphs are underutilized and lack correlation mining. Existing methods struggle to effectively extract structural knowledge between policies and semantic connections between enterprise data. The automation level is low, making it difficult to achieve real-time dynamic matching and report generation.

[0004] Therefore, there is an urgent need for a data matching method and system that integrates AI models and knowledge graph technology, and has the capabilities of multi-source data fusion, deep semantic understanding and intelligent matching, in order to improve the intelligence and accuracy of policy application. Summary of the Invention

[0005] To address the problems of existing technologies, such as single data sources, incomplete information, crude matching mechanisms, lack of semantic understanding, insufficient utilization of knowledge graphs, lack of correlation mining, and low automation, the present invention aims to provide a data matching method and system driven by AI models and knowledge graphs. This method and system can improve the intelligence level of policy matching, provide enterprises with personalized and high-precision policy recommendation support, and enhance the scientific and effective allocation of government resources. It has good practical value and promising prospects for promotion.

[0006] To solve the above problems, the present invention adopts the following technical solution.

[0007] A data matching method based on AI models and knowledge graphs, comprising the following steps: It intelligently captures policy data and collects enterprise data, and performs document processing and structured storage. Constructing a policy knowledge graph and enhancing enterprise data, including knowledge base initialization and AI-assisted modeling; The system integrates a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. Generate a dynamic report, which includes matching results and application recommendations; The execution steps of the causal discovery algorithm module include: Construct a policy-behavior-environment data cube, integrating policy texts, behavioral logs, sentiment data, and unstructured information; Graph neural networks are used to generate causal graphs, and policy shock simulation and counterfactual reasoning are introduced. Identify confounding factors and use backdoor adjustments and black hole detection to track policy failure paths; Conduct policy mutation experiments using digital twins to simulate the evolution of game strategies; Establish a causal strength index and compress key causal dimensions by combining semantic similarity and information bottlenecks; By introducing feedback mechanisms and anomaly alert systems, and leveraging blockchain technology, the causal process can be traced.

[0008] Furthermore, policy data collection includes the following steps: The URL parser identifies the structure of the policy website and generates targeted data collection paths. The web crawler engine is used to collect data by category and extract key information, including policy title, release date, and text link. The policy attachments were converted to Markdown format for subsequent AI processing. The processed policy data is stored in a structured database and indexed to support fast retrieval.

[0009] Furthermore, the enterprise data collection adopts a dual-track mechanism, including the following steps: Collect basic information from businesses through a front-end form system; The AI ​​search engine is used to proactively obtain publicly available information from enterprises, including business registration data, industry trends, and qualification certifications. By merging form data and AI search data through a data fusion processor, data conflicts are resolved and high-quality enterprise profiles are generated; The integrity of the data is verified by a quality verifier to ensure that the enterprise information dimensions are expanded to include business registration data and operating status.

[0010] Furthermore, the knowledge base initialization includes the following steps: Import policy data into the knowledge base using a data migration engine; Analyze the characteristics of policy data to identify key entity types and relationship patterns; The application conditions and qualification requirements are extracted from policy texts using a rule pattern recognizer and stored in the rule base. A policy knowledge graph is built based on prompt words, supporting dynamic updates and expansion; By using an entity alignment engine, business registration information and qualification data in enterprise files are mapped to a knowledge graph, establishing a two-way link between policy and enterprise.

[0011] Furthermore, AI-assisted knowledge modeling includes the following steps: The policy document is segmented into independent semantic units using a semantic text segmenter; The vectorization engine converts semantic units into high-dimensional vectors, capturing deep semantic information; Feature extractors identify entity types, attribute features, and relationship patterns; The graph builder creates knowledge graph nodes and edges, while the relationship reasoning engine uncovers hidden connections.

[0012] Furthermore, the hybrid matching engine includes the following steps: The enterprise feature scanner dynamically extracts multi-dimensional enterprise features, which include static attributes and operating status. The context builder embeds enterprise characteristics and policy conditions into a unified semantic space; The similarity calculation engine combines vector similarity with rule matching to generate hybrid matching results; The result fusion module combines the outputs of the two mechanisms to generate the final matching score and confidence level.

[0013] Furthermore, the vector similarity calculation employs a semantic encoder based on the Transformer architecture, combined with contrastive learning to optimize the vector representation. The calculation formula is as follows:

[0014] in, S represents the overall matching score between policies and enterprises; The cosine similarity between the policy embedding vector and the enterprise embedding vector; The text semantic matching score is based on a pre-trained model; For domain-adaptive adjustment functions; , , For learnable weight parameters, satisfying + + =1.0.

[0015] Furthermore, dynamic report generation includes the following steps: The global query engine mines the complete information chain of matching policies; The report template engine allows you to choose between preset templates or personalized customization. Content generators convert structured data into easily readable text for description; The formatter supports exporting in PDF, Word, and HTML formats.

[0016] Furthermore, knowledge base initialization also includes the following steps: Establish an incremental update mechanism to monitor changes in policy and enterprise data through change detectors, triggering partial updates to the knowledge graph.

[0017] A data matching system driven by AI models and knowledge graphs, which applies the aforementioned data matching method driven by AI models and knowledge graphs, includes the following units: The data acquisition unit is used to intelligently capture policy data and collect enterprise data, and to perform document processing and structured storage. The data processing unit is used to build a policy knowledge graph and enhance enterprise data, including knowledge base initialization and AI-assisted modeling. The policy matching unit is used to integrate a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. The report generation unit is used to generate dynamic reports, which include matching results and application suggestions.

[0018] The advantages of this invention are: 1. By intelligently capturing policy data and enterprise information, the entire process is automated, from data collection to matching and recommendation, without the need for manual intervention.

[0019] 2. A dual-track mechanism is adopted to collect enterprise information, integrating self-reported data from enterprises with publicly available data obtained through AI search, to generate high-quality enterprise profiles.

[0020] 3. Construct a policy knowledge graph and combine it with AI-assisted modeling to achieve policy semantic structuring and condition rule-based approach, thereby improving the semantic depth of matching.

[0021] 4. Integrate vector similarity and rule matching mechanisms, and combine semantic understanding and logical reasoning to improve the accuracy of matching policies with enterprises.

[0022] 5. The dynamic report generation module outputs matching scores, matching criteria, and application suggestions to facilitate enterprise decision-making.

[0023] 6. Establish an incremental update process to ensure that the policy knowledge graph and enterprise data are updated synchronously, adapting to the dual dynamics of policy changes and enterprise growth. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1: Please see Figure 1 This invention provides a technical solution: a data matching method driven by AI models and knowledge graphs, the method comprising the following steps: It intelligently captures policy data and collects enterprise data, and performs document processing and structured storage. Intelligent crawling refers to the use of AI-driven web crawling technology, combined with natural language processing (NLP) and computer vision (CV), to automatically identify the structure of policy websites and collect information in a targeted manner. Policy data refers to normative documents issued by governments or institutions, including structured data such as policy titles, issuing authorities, release dates, and validity periods; and unstructured data such as policy text, attachments (e.g., application guidelines, flowcharts), and links to related regulations. Enterprise data collection is achieved through a two-track mechanism: passive collection, where enterprises fill in basic information (e.g., name, registered address, industry classification) through front-end forms; and active collection, which utilizes AI search engines to crawl publicly available data (e.g., business registration information, patent layout, and public opinion dynamics). Document processing and structured storage involve cleaning, transforming, and storing the collected raw data. Constructing a policy knowledge graph and enhancing enterprise data, including knowledge base initialization and AI-assisted modeling; Among them, the policy knowledge graph is a database that stores policy knowledge in a graph structure, including policy name, issuing authority, application conditions, and subsidy standards; enterprise data enhancement expands the dimensions of enterprise information through the knowledge graph: knowledge base initialization is the initial construction stage of the knowledge graph, including: data migration: importing structured policy data into a graph database (such as Neo4j); rule extraction: extracting structured rules such as application conditions and prohibitions from policy text; AI-assisted modeling is to automatically construct the knowledge graph using machine learning technology: semantic segmentation: splitting policy documents into clause-level semantic units; vectorization: converting text into numerical vectors to capture deep semantic relationships; The system integrates a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. Among them, the hybrid matching engine is a decision-making system that combines two matching strategies: vector similarity matching, which calculates the semantic similarity between enterprise characteristics and policy conditions; rule-based hard matching, which strictly verifies whether enterprises meet the rigid conditions in the policy; rule-based reasoning mechanism, which is a decision-making system based on logical rules; and dynamic precise matching, which is the ability to adjust the matching strategy in real time. Generate a dynamic report, which includes matching results and application recommendations; Among them, the dynamic report is a personalized document generated in real time based on the matching results.

[0027] The execution steps of the causal discovery algorithm module include: Construct a policy-behavior-environment data cube, integrating policy texts, behavioral logs, sentiment data, and unstructured information; Graph neural networks are used to generate causal graphs, and policy shock simulation and counterfactual reasoning are introduced. Identify confounding factors and use backdoor adjustments and black hole detection to track policy failure paths; Conduct policy mutation experiments using digital twins to simulate the evolution of game strategies; Establish a causal strength index and compress key causal dimensions by combining semantic similarity and information bottlenecks; Introduce feedback mechanisms and anomaly alert systems, and leverage blockchain to ensure the traceability of causal processes; Among them, through the complete technical link of data fusion → causal modeling → hybrid control → strategy simulation → credible verification, a closed-loop causal tracking from policy release to enterprise response is realized, providing a full-process solution for intelligent policy matching from causal discovery to credible verification, which is especially suitable for precise policy implementation and enterprise compliance declaration scenarios in complex policy environments.

[0028] It should be noted that during operation, the data layer provides raw materials for the knowledge graph, structured storage ensures data retrieval, the knowledge graph provides decision-making basis for the matching engine, AI modeling improves graph coverage, hybrid matching combines rule-based hard matching and semantic soft matching, strategies are dynamically adjusted to adapt to policy / enterprise changes, dynamic reports translate technical results into business language, and template engines and global queries support decision-making transparency.

[0029] Specifically, policy data collection includes the following steps: The URL parser identifies the structure of the policy website and generates targeted data collection paths. The web crawler engine is used to collect data by category and extract key information, including policy title, release date, and text link. The policy attachments were converted to Markdown format for subsequent AI processing. The processed policy data is stored in a structured database and indexed to support fast retrieval.

[0030] This design automatically identifies the policy website's structure using a URL parser, generating targeted data collection paths to ensure the crawler accurately retrieves target data. The crawler engine extracts metadata such as policy titles and publication dates by category, and performs OCR recognition and Markdown conversion on attachments (PDF / Word) before storing them in a structured database and creating an index. Targeted paths avoid crawling invalid data, improving efficiency by over 30%; Markdown conversion allows unstructured attachments to be directly parsed by AI, reducing subsequent processing complexity; and structured storage and indexing technology support millisecond-level keyword queries, meeting the needs of real-time policy monitoring.

[0031] Specifically, enterprise data collection adopts a dual-track mechanism, including the following steps: Collect basic information from businesses through a front-end form system; The AI ​​search engine is used to proactively obtain publicly available information from enterprises, including business registration data, industry trends, and qualification certifications. By merging form data and AI search data through a data fusion processor, data conflicts are resolved and high-quality enterprise profiles are generated; The integrity of the data is verified by a quality verifier to ensure that the enterprise information dimensions are expanded to include business registration data and operating status.

[0032] This design employs a dual-track mechanism, combining front-end forms with an AI search engine. A data fusion processor resolves conflicts, generating high-quality profiles, which are then verified for completeness by a quality validator. It integrates proactive and reactive data collection, covering both static enterprise attributes (such as qualifications) and dynamic operational data (such as financial reports). AI algorithms automatically identify and correct data inconsistencies (such as discrepancies between the registered business address and the submitted location), enhancing the profile's credibility. Furthermore, it extends from basic information to industry benchmarking data (such as policy matching rates for similar enterprises), providing a basis for accurate matching.

[0033] Specifically, knowledge base initialization includes the following steps: Import policy data into the knowledge base using a data migration engine; Analyze the characteristics of policy data to identify key entity types and relationship patterns; The application conditions and qualification requirements are extracted from policy texts using a rule pattern recognizer and stored in the rule base. A policy knowledge graph is built based on prompt words, supporting dynamic updates and expansion; By using an entity alignment engine, business registration information and qualification data in enterprise files are mapped to a knowledge graph, establishing a two-way link between policy and enterprise.

[0034] This design utilizes a data migration engine to import policy data into a knowledge base, analyzes features to identify entities and relationships, extracts application rules and stores them in a rule base, and an entity alignment engine to map enterprise data to a knowledge graph, establishing a two-way relationship between policy and enterprises and supporting incremental updates. Policy text is transformed into graph nodes / edges, improving semantic relevance; extracted application conditions can be directly used for matching and verification; and an incremental update mechanism (such as monitoring policy revisions) ensures the timeliness of the knowledge graph and avoids the resource consumption of full reconstruction.

[0035] Specifically, AI-assisted knowledge modeling includes the following steps: The policy document is segmented into independent semantic units using a semantic text segmenter; The vectorization engine converts semantic units into high-dimensional vectors, capturing deep semantic information; Feature extractors identify entity types, attribute features, and relationship patterns; The graph builder creates knowledge graph nodes and edges, while the relationship reasoning engine uncovers hidden connections.

[0036] This design allows the semantic segmenter to break down policy documents into clause-level units, the vectorization engine to convert them into high-dimensional vectors to capture deep semantics, the feature extractor to identify entity types, the graph builder to create nodes / edges, and the relationship reasoning engine to uncover implicit connections (such as "Policy A → Indirect Support → Industrial Chain C"). Clause-level processing improves matching accuracy and avoids interference from noise throughout the document; it captures implicit conditions such as "High-tech Enterprise Certification Must Possess Independent Intellectual Property Rights"; and it discovers synergistic relationships between policies (such as the application path of "Environmental Policy A + Tax Incentive B" combination).

[0037] Specifically, the hybrid matching engine includes the following steps: The enterprise feature scanner dynamically extracts multi-dimensional enterprise features, which include static attributes and operating status. The context builder embeds enterprise characteristics and policy conditions into a unified semantic space; The similarity calculation engine combines vector similarity with rule matching to generate hybrid matching results; The result fusion module combines the outputs of the two mechanisms to generate the final matching score and confidence level.

[0038] This design allows the enterprise feature scanner to dynamically extract static attributes (such as registered location) and operational data (such as revenue). The context builder embeds features and policy conditions into the same semantic space. The similarity engine combines vector matching (semantic similarity) and rule matching (such as "must be a local enterprise"), and the results fusion module outputs the final score. Rule matching ensures that rigid conditions (such as registered location) are not overlooked, while vector matching captures flexible conditions (such as technological relevance). Updates to enterprise data (such as new patents) automatically trigger rematching, avoiding lag in static matching. The fused score provides a reference for matching credibility, assisting in decision-making and prioritization.

[0039] Specifically, the vector similarity calculation employs a semantic encoder based on the Transformer architecture, combined with contrastive learning to optimize the vector representation. The calculation formula is as follows:

[0040] in, S represents the overall matching score between policies and enterprises; The cosine similarity between the policy embedding vector and the enterprise embedding vector; The text semantic matching score is based on a pre-trained model; A domain-adaptive adjustment function is used to dynamically balance the weights of general features and industry-specific features. , , For learnable weight parameters, satisfying + + =1.0, optimized through multi-task learning.

[0041] This design uses a multimodal fusion calculation to obtain the final similarity index, which is then used to quantify the degree of matching between policy provisions and corporate characteristics.

[0042] Specifically, dynamic report generation includes the following steps: The global query engine mines the complete information chain of matching policies; The report template engine allows you to choose between preset templates or personalized customization. Content generators convert structured data into easily readable text for description; The formatter supports exporting in PDF, Word, and HTML formats.

[0043] This design allows the global query engine to trace and match the chain of evidence, the report template engine to support preset formats or personalized customization, the content generator to convert structured data into natural language descriptions, and the formatter to export PDF / Word / HTML. The evidence chain is labeled with matching criteria, increasing user trust in the recommended results; template customization meets the needs of different scenarios (such as executive briefings vs. application materials); and formatted export supports multiple applications such as printing, email attachments, and web page display.

[0044] Specifically, knowledge base initialization also includes the following steps: Establish an incremental update mechanism to monitor changes in policy and enterprise data through change detectors, triggering partial updates to the knowledge graph.

[0045] This design changes the monitors policy data (such as clause revisions) and enterprise data (such as business registration changes), triggering partial updates to the knowledge graph (such as modifying the attributes of associated nodes). The graph update is completed within one hour after the policy revision, ensuring the timeliness of matching results; partial updates save 80% of computing resources compared to full reconstruction; update history is recorded, supporting rollback to any version at any point in time, improving system traceability.

[0046] Example 2: Please see Figure 2 , A data matching system based on AI models and knowledge graphs, applicable to the aforementioned data matching method based on AI models and knowledge graphs, comprises the following units: The data acquisition unit is used to intelligently capture policy data and collect enterprise data, and to perform document processing and structured storage. The data processing unit is used to build a policy knowledge graph and enhance enterprise data, including knowledge base initialization and AI-assisted modeling. The policy matching unit is used to integrate a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. The report generation unit is used to generate dynamic reports, which include matching results and application suggestions.

[0047] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A data matching method based on AI models and knowledge graphs, characterized in that, The method includes the following steps: It intelligently captures policy data and collects enterprise data, and performs document processing and structured storage. Constructing a policy knowledge graph and enhancing enterprise data, including knowledge base initialization and AI-assisted modeling; The system integrates a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. Generate a dynamic report, which includes matching results and application recommendations; The execution steps of the causal discovery algorithm module include: Construct a policy-behavior-environment data cube, integrating policy texts, behavioral logs, sentiment data, and unstructured information; Graph neural networks are used to generate causal graphs, and policy shock simulation and counterfactual reasoning are introduced. Identify confounding factors and use backdoor adjustments and black hole detection to track policy failure paths; Conduct policy mutation experiments using digital twins to simulate the evolution of game strategies; Establish a causal strength index and compress key causal dimensions by combining semantic similarity and information bottlenecks; By introducing feedback mechanisms and anomaly alert systems, and leveraging blockchain technology, the causal process can be traced.

2. The data matching method based on AI model and knowledge graph driven according to claim 1, characterized in that, Policy data collection includes the following steps: The URL parser identifies the structure of the policy website and generates targeted data collection paths. The web crawler engine is used to collect data by category and extract key information, including policy title, release date, and text link. The policy attachments were converted to Markdown format for subsequent AI processing. The processed policy data is stored in a structured database and indexed to support fast retrieval.

3. The data matching method based on AI model and knowledge graph driven according to claim 2, characterized in that, Enterprise data collection adopts a dual-track mechanism, including the following steps: Collect basic information from businesses through a front-end form system; The AI ​​search engine is used to proactively obtain publicly available information from enterprises, including business registration data, industry trends, and qualification certifications. By merging form data and AI search data through a data fusion processor, data conflicts are resolved and high-quality enterprise profiles are generated; The integrity of the data is verified by a quality verifier to ensure that the enterprise information dimensions are expanded to include business registration data and operating status.

4. The data matching method based on AI model and knowledge graph driven according to claim 3, characterized in that, Knowledge base initialization includes the following steps: Import policy data into the knowledge base using a data migration engine; Analyze the characteristics of policy data to identify key entity types and relationship patterns; The application conditions and qualification requirements are extracted from policy texts using a rule pattern recognizer and stored in the rule base. A policy knowledge graph is built based on prompt words, supporting dynamic updates and expansion; By using an entity alignment engine, business registration information and qualification data in enterprise files are mapped to a knowledge graph, establishing a two-way link between policy and enterprise.

5. The data matching method based on AI model and knowledge graph driven according to claim 4, characterized in that, AI-assisted knowledge modeling includes the following steps: The policy document is segmented into independent semantic units using a semantic text segmenter; The vectorization engine converts semantic units into high-dimensional vectors, capturing deep semantic information; Feature extractors identify entity types, attribute features, and relationship patterns; The graph builder creates knowledge graph nodes and edges, while the relationship reasoning engine uncovers hidden connections.

6. The data matching method based on AI model and knowledge graph driven according to claim 5, characterized in that, The hybrid matching engine includes the following steps: The enterprise feature scanner dynamically extracts multi-dimensional enterprise features, which include static attributes and operating status. The context builder embeds enterprise characteristics and policy conditions into a unified semantic space; The similarity calculation engine combines vector similarity with rule matching to generate hybrid matching results; The result fusion module combines the outputs of the two mechanisms to generate the final matching score and confidence level.

7. The data matching method based on AI model and knowledge graph driven according to claim 6, characterized in that, The vector similarity calculation employs a semantic encoder based on the Transformer architecture, combined with contrastive learning to optimize the vector representation. The calculation formula is as follows: in, S represents the overall matching score between policies and enterprises; The cosine similarity between the policy embedding vector and the enterprise embedding vector; The text semantic matching score is based on a pre-trained model; For domain-adaptive adjustment functions; , , For learnable weight parameters, satisfying + + =1.

0.

8. The data matching method based on AI model and knowledge graph driven according to claim 7, characterized in that, Dynamic report generation includes the following steps: The global query engine mines the complete information chain of matching policies; The report template engine allows you to choose between preset templates or personalized customization. Content generators convert structured data into easily readable text for description; The formatter supports exporting in PDF, Word, and HTML formats.

9. The data matching method based on AI model and knowledge graph driven according to claim 4, characterized in that, Knowledge base initialization also includes the following steps: Establish an incremental update mechanism to monitor changes in policy and enterprise data through change detectors, triggering partial updates to the knowledge graph.

10. A data matching system driven by AI models and knowledge graphs, applicable to the data matching method driven by AI models and knowledge graphs as described in any one of claims 1-9, characterized in that, The system includes the following units: The data acquisition unit is used to intelligently capture policy data and collect enterprise data, and to perform document processing and structured storage. The data processing unit is used to build a policy knowledge graph and enhance enterprise data, including knowledge base initialization and AI-assisted modeling. The policy matching unit is used to integrate a hybrid matching engine and a rule-based reasoning mechanism to achieve dynamic and accurate matching between enterprises and policies. The hybrid matching engine embeds a causal discovery algorithm module to identify the causal relationship between policy provisions and enterprise behavior, thereby avoiding false matching caused by spurious correlations. The report generation unit is used to generate dynamic reports, which include matching results and application suggestions.

Images