Policy tagging method and system based on large model technology

Abstract

The invention discloses a policy tagging method and system based on a large model technology, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting policy texts from a government website and a policy database, and carrying out the cleaning, word segmentation and entity recognition; training a policy semantic understanding model based on a Transform architecture, and optimizing the long-tail semantic recognition capability through comparative learning; policy knowledge graph construction: storing policy entities, labels and association relationships thereof, and supporting dynamic updating and semantic reasoning; generating candidate tags by using the large model, and performing semantic matching and priority ranking in combination with the knowledge graph; and providing label auditing, correcting and feedback functions, and optimizing the model through Active Learning. According to the method, the problems of low efficiency, poor accuracy, dependence on manual intervention and the like in the traditional policy text tagging process are solved, the efficiency and the accuracy can be improved, the semantic comprehension capability is enhanced, and dynamic extension is supported.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and specifically to a method and system for policy tagging based on large model technology. Background Art

[0002] The following problems mainly exist in traditional policy text tagging:

[0003] Manual annotation method: Traditional policy tagging relies on manual reading and annotation, which is inefficient and vulnerable to subjective factors (such as inconsistent annotation standards). Keyword matching technology: Rule-based matching methods based on keywords (such as regular expressions, TF-IDF) are difficult to capture the deep semantics of policy texts, resulting in inaccurate label classification. Traditional machine learning models: Such as SVM, random forest, etc., rely on a large amount of manual feature engineering and have insufficient generalization ability for new policy texts.

[0004] In summary, traditional policy text tagging has the following problems: low efficiency: Manual annotation is time-consuming and laborious, and cannot cope with the rapid update of a large amount of policy texts; insufficient semantic understanding: Traditional methods cannot parse the complex semantic relationships implicit in policy texts (such as nested conditions of policy application objects); poor dynamic adaptability: The label system is rigid and difficult to adapt to policy content changes and user-defined requirements. Summary of the Invention

[0005] The technical task of the present invention is to address the above deficiencies, providing a method and system for policy tagging based on large model technology, which solves problems such as low efficiency, poor accuracy, and dependence on manual intervention in the process of traditional policy text tagging, can improve efficiency and accuracy, enhance semantic understanding ability, and support dynamic expansion, and is applicable to intelligent classification, tagging, and semantic association analysis of policy texts.

[0006] The technical solution adopted by the present invention to solve its technical problems is:

[0007] A method for policy tagging based on large model technology, the implementation of which includes:

[0008] Data collection and preprocessing: Collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition;

[0009] Large model training and optimization: Train a policy semantic understanding model based on the Transformer architecture, and optimize the long-tail semantic recognition ability through contrastive learning;

[0010] Policy knowledge graph construction: Store policy entities, tags, and their association relationships, and support dynamic update and semantic reasoning;

[0011] Label Generation and Matching: Use large models to generate candidate labels, and combine with knowledge graphs for semantic matching and priority ranking;

[0012] User Interaction and Feedback: Provide label review, correction, and feedback functions, and optimize the model through Active Learning.

[0013] This method uses large model technology to conduct in-depth semantic understanding of policy texts, combines policy knowledge graphs and dynamic label generation algorithms to achieve intelligent classification, multi-label generation, and semantic association of policy texts.

[0014] Furthermore, the data collection and preprocessing are specifically implemented as follows:

[0015] Crawl policy texts from channels including government websites, public databases, and user manual input, and perform cleaning, denoising, and structured storage;

[0016] Adopt distributed crawler technology to support incremental updates;

[0017] Preprocess the text, including word segmentation, entity recognition (such as "declaration entity", "support amount", etc.).

[0018] Furthermore, the large model training and optimization specifically include:

[0019] Train a dedicated large model based on a large number of policy texts to optimize semantic understanding ability;

[0020] Use the Transformer architecture for pre-training and perform domain adaptation on policy texts;

[0021] Introduce contrastive learning to enhance the model's ability to recognize long-tail semantics in policy texts.

[0022] Furthermore, the construction of the policy knowledge graph specifically includes:

[0023] Build a structured knowledge base to store policy entities (such as "high-tech enterprises", "tax incentives"), labels, and association relationships;

[0024] Extract entities and relationships from policy texts through large models to automatically expand the knowledge graph.

[0025] Furthermore, the policy knowledge graph is implemented based on the Neo4j graph database, supporting dynamic updates and semantic reasoning.

[0026] Furthermore, the label generation and matching specifically include:

[0027] Use large models to generate candidate labels, and perform semantic matching and priority ranking through knowledge graphs;

[0028] Generate candidate labels (such as "Technological Innovation" and "Support for Small, Medium and Micro Enterprises") using Few-shot Learning;

[0029] Calculate the semantic similarity between the labels and the policy text based on the knowledge graph, and screen the optimal label combination.

[0030] Furthermore, the user interaction and feedback are specifically implemented as follows:

[0031] Provide a visual interface for users to review the labels, correct errors, and collect feedback data to optimize the model;

[0032] Support users to customize the label system, and select high-value samples through the Active Learning strategy to iteratively train the model.

[0033] The present invention also claims to protect a policy label system based on large model technology, including:

[0034] A data collection and preprocessing module, which is used to collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition;

[0035] A large model training and optimization module, which trains a policy semantic understanding model based on the Transformer architecture, and optimizes the long-tail semantic recognition ability through contrastive learning;

[0036] A policy knowledge graph construction module, which is used to store policy entities, labels and their association relationships, and support dynamic update and semantic reasoning;

[0037] A label generation and matching module, which uses the large model to generate candidate labels, and combines the knowledge graph for semantic matching and priority ranking;

[0038] A user interaction and feedback module, which provides label review, correction and feedback functions, and optimizes the model through Active Learning;

[0039] This system can implement the above-mentioned policy labeling method based on large model technology.

[0040] The present invention also claims to protect a policy labeling device based on large model technology, including: at least one memory and at least one processor;

[0041] The at least one memory is used to store machine-readable programs;

[0042] The at least one processor is used to call the machine-readable program to implement the above method.

[0043] The present invention also claims protection for a computer-readable medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the above-mentioned method is implemented.

[0044] Compared with the prior art, the policy tagging method and system based on large model technology of the present invention have the following beneficial effects:

[0045] 1. Efficiency improvement: The speed of automatic tag generation is more than 10 times faster than manual annotation, and it supports real-time processing of new policy texts.

[0046] 2. Accuracy enhancement: On the test set, the accuracy rate of tag classification reaches 95% (compared with 78% of the traditional method).

[0047] 3. High flexibility: Users can customize the tag hierarchy (such as "industry-region-policy type") to meet diverse needs.

[0048] 4. Strong interpretability: The tag generation logic is visually displayed through a knowledge graph to assist users in understanding the decision-making process. Description of the Drawings

[0049] Figure 1 It is a schematic diagram of the policy tagging system architecture based on large model technology provided by an embodiment of the present invention. Detailed Embodiments

[0050] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0051] An embodiment of the present invention provides a policy tagging method based on large model technology, and the implementation of this method includes:

[0052] Data collection and preprocessing: Collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition.

[0053] Large model training and optimization: Train a policy semantic understanding model based on the Transformer architecture, and optimize the long-tail semantic recognition ability through contrastive learning.

[0054] Policy knowledge graph construction: Store policy entities, tags, and their association relationships, and support dynamic update and semantic reasoning.

[0055] Tag generation and matching: Use the large model to generate candidate tags, and perform semantic matching and priority sorting in combination with the knowledge graph.

[0056] User interaction and feedback: Provide functions for tag review, correction, and feedback, and optimize the model through Active Learning.

[0057] This method conducts in-depth semantic understanding of policy texts through large model technology, combines a policy knowledge graph and a dynamic label generation algorithm to achieve intelligent classification, multi-label generation, and semantic association of policy texts. It includes a data collection and preprocessing module, a large model training and optimization module, a policy knowledge graph construction module, a label generation and matching module, and a user interaction and feedback module. It can be widely applied in fields such as policy management, enterprise services, data analysis, etc., significantly improving the efficiency and accuracy of policy labeling.

[0058] The specific implementation of this method is as follows:

[0059] 1. Data collection and preprocessing:

[0060] Crawl policy texts from channels such as government websites, public databases, and user manual input, and perform cleaning, denoising, and structured storage.

[0061] Adopt distributed crawler technology to support incremental updates. Perform preprocessing such as word segmentation and entity recognition (such as "declaration entity", "support amount") on the text.

[0062] 2. Large model training and optimization:

[0063] Train a dedicated large model based on a large amount of policy texts to optimize semantic understanding ability.

[0064] Use the Transformer architecture for pre-training and perform domain adaptation on policy texts.

[0065] Introduce contrastive learning to enhance the model's ability to recognize long-tail semantics in policy texts.

[0066] 3. Policy knowledge graph construction:

[0067] Construct a structured knowledge base to store policy entities (such as "high-tech enterprises", "tax incentives"), labels, and association relationships.

[0068] Implemented based on the Neo4j graph database, supporting dynamic updates and semantic reasoning.

[0069] Extract entities and relationships from policy texts through a large model to automatically expand the knowledge graph.

[0070] 4. Label generation and matching:

[0071] Construct a structured knowledge base to store policy entities (such as "high-tech enterprises", "tax incentives"), labels, and association relationships.

[0072] Implemented based on the Neo4j graph database, supporting dynamic updates and semantic reasoning. Extract entities and relationships from policy texts through a large model to automatically expand the knowledge graph.

[0073] 5. User Interaction and Feedback:

[0074] Provide a visual interface for users to review tags, correct errors, and collect feedback data to optimize the model.

[0075] Support users to customize the tag system and select high-value samples through the Active Learning strategy to iteratively train the model.

[0076] This method uses a large model to automatically parse the semantics of policy texts, generate accurate tags, and improve efficiency and accuracy. Combining with a policy knowledge graph, it identifies entities, relationships, and implicit logics in the text to enhance semantic understanding ability. It supports dynamic expansion, allows users to customize the tag system, and continuously optimizes the tag generation algorithm through a feedback mechanism. It solves the problems of low efficiency, poor accuracy, and dependence on manual intervention in the traditional policy text tagging process.

[0077] The embodiment of the present invention also provides a policy tagging system based on large model technology, including:

[0078] A data collection and preprocessing module, which is used to collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition;

[0079] A large model training and optimization module, which trains a policy semantic understanding model based on the Transformer architecture and optimizes the long-tail semantic recognition ability through contrastive learning;

[0080] A policy knowledge graph construction module, which is used to store policy entities, tags, and their association relationships, and support dynamic update and semantic reasoning;

[0081] A tag generation and matching module, which uses a large model to generate candidate tags, and performs semantic matching and priority ranking in combination with the knowledge graph;

[0082] A user interaction and feedback module, which provides functions such as tag review, correction, and feedback, and optimizes the model through Active Learning;

[0083] This system can implement the policy tagging method based on large model technology described in the above embodiment.

[0084] The data collection and preprocessing module,

[0085] Crawl policy texts from channels such as government websites, public databases, and user manual input, and perform cleaning, denoising, and structured storage.

[0086] Adopt distributed crawler technology to support incremental updates. Perform preprocessing such as word segmentation and entity recognition (such as "declaration entity", "support amount") on the text.

[0087] The large model training and optimization module,

[0088] Train a dedicated large model based on a large amount of policy texts to optimize semantic understanding ability.

[0089] Perform pre-training using the Transformer architecture and adapt to the domain on policy texts.

[0090] Introduce contrastive learning to enhance the model's ability to recognize long-tail semantics in policy texts.

[0091] The policy knowledge graph construction module

[0092] Construct a structured knowledge base to store policy entities (such as "high-tech enterprises", "tax incentives"), labels, and association relationships.

[0093] Implemented based on the Neo4j graph database, supporting dynamic updates and semantic reasoning.

[0094] Extract entities and relationships from policy texts through the large model to automatically expand the knowledge graph.

[0095] The label generation and matching module

[0096] Construct a structured knowledge base to store policy entities (such as "high-tech enterprises", "tax incentives"), labels, and association relationships.

[0097] Implemented based on the Neo4j graph database, supporting dynamic updates and semantic reasoning. Extract entities and relationships from policy texts through the large model to automatically expand the knowledge graph.

[0098] The user interaction and feedback module

[0099] Provide a visual interface for users to review labels, correct errors, and collect feedback data to optimize the model.

[0100] Support users to customize the label system and select high-value samples through the Active Learning strategy to iteratively train the model.

[0101] An embodiment of the present invention also provides a policy tagging device based on large model technology, including: at least one memory and at least one processor;

[0102] The at least one memory is used to store machine-readable programs;

[0103] The at least one processor is used to call the machine-readable program to implement the policy tagging method based on large model technology described in the above embodiments.

[0104] An embodiment of the present invention further provides a computer-readable medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the processor is caused to execute the policy tagging method based on large model technology described in the above embodiments. Specifically, a system or device equipped with a storage medium can be provided, on which software program code for implementing the functions of any one of the above embodiments is stored, and the computer (or CPU or MPU) of the system or device is caused to read and execute the program code stored in the storage medium.

[0105] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments. Therefore, the program code and the storage medium storing the program code constitute a part of the present invention.

[0106] Examples of the storage medium for providing the program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code can be downloaded from a server computer via a communication network.

[0107] In addition, it should be clear that not only can the functions of any one of the above embodiments be realized by executing the program code read by the computer, but also by causing an operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.

[0108] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or the memory provided in the expansion unit connected to the computer, and then based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion unit is caused to execute part and all of the actual operations, thereby realizing the functions of any one of the above embodiments.

[0109] The present invention has been described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above-mentioned multiple embodiments, those skilled in the art can know that more embodiments of the present invention can be obtained by combining the code review means in the above different embodiments, and these embodiments are also within the protection scope of the present invention.

Claims

1. A policy tagging method based on large model technology, characterized in that, The implementation of this method includes: Data collection and preprocessing: Collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition; Large model training and optimization: Train a policy semantic understanding model based on the Transformer architecture, and optimize the long-tail semantic recognition ability through contrastive learning; Policy knowledge graph construction: Store policy entities, labels, and their association relationships, and support dynamic updates and semantic reasoning; Label generation and matching: Use the large model to generate candidate labels, and perform semantic matching and priority ranking in combination with the knowledge graph; User interaction and feedback: Provide functions for label review, correction, and feedback, and optimize the model through Active Learning.

2. The policy tagging method based on large model technology according to claim 1, wherein, The specific implementation of the data collection and preprocessing is as follows: Crawl policy texts, and the channels include government websites, public databases, and manual user input, and perform cleaning, denoising, and structured storage; Adopt distributed crawler technology to support incremental updates; Preprocess the text, including word segmentation, entity recognition (such as "declaration entity", "support amount", etc.).

3. A method for policy tagging based on large model technology according to claim 1, characterized in that The specific implementation of the large model training and optimization includes: Train a dedicated large model based on a large number of policy texts to optimize the semantic understanding ability; Perform pre-training using the Transformer architecture and perform domain adaptation on policy texts; Introduce contrastive learning to enhance the model's ability to recognize long-tail semantics in policy texts.

4. A method for policy tagging based on large model technology according to claim 1, characterized in that, The specific implementation of the policy knowledge graph construction includes: Construct a structured knowledge base to store policy entities, labels, and association relationships; Extract entities and relationships from policy texts through the large model to automatically expand the knowledge graph.

5. A method for policy tagging based on large model technology according to claim 1 or 4, characterized in that The policy knowledge graph is implemented based on the Neo4j graph database, and supports dynamic updates and semantic reasoning.

6. A method for policy tagging based on large model technology according to claim 1, characterized in that The specific implementation of the label generation and matching includes: Use the large model to generate candidate labels, and perform semantic matching and priority ranking through the knowledge graph; Adopt Few-shot Learning to generate candidate labels; Calculate the semantic similarity between labels and policy texts based on the knowledge graph, and screen the optimal label combination.

7. A method for policy tagging based on large model technology according to claim 1, characterized in that The specific implementation of the user interaction and feedback includes: Provide a visual interface for users to review labels, correct errors, and collect feedback data to optimize the model; Support users to customize the label system, and select high-value samples through the Active Learning strategy to iteratively train the model.

8. A policy tagging system based on large model technology, characterized in that, It includes: A data collection and preprocessing module, which is used to collect policy texts from government websites and policy databases, and perform cleaning, word segmentation, and entity recognition; A large model training and optimization module, which trains a policy semantic understanding model based on the Transformer architecture, and optimizes the long-tail semantic recognition ability through contrastive learning; A policy knowledge graph construction module, which is used to store policy entities, labels, and their association relationships, and support dynamic updates and semantic reasoning; A label generation and matching module, which uses the large model to generate candidate labels, and performs semantic matching and priority ranking in combination with the knowledge graph; A user interaction and feedback module, which provides functions for label review, correction, and feedback, and optimizes the model through Active Learning; The system can implement the policy tagging method based on large model technology described in any one of claims 1 to 7.

9. A policy tagging device based on large model technology, characterized in that, It includes: At least one memory and at least one processor; The at least one memory is used to store machine-readable programs; The at least one processor is used to call the machine-readable program to implement the method described in any one of claims 1 to 7.

10. A computer-readable medium, characterized in that, Computer instructions are stored on the computer-readable medium, and when the computer instructions are executed by the processor, the method described in any one of claims 1 to 7 is implemented.

Images