Policy traceability dynamic retrieval method based on multilevel intention recognition and comparative learning

Abstract

The invention discloses a policy traceability dynamic retrieval method based on multi-level intention recognition and comparative learning, relates to the technical field of natural language processing and information retrieval and generation, and introduces multi-level intention recognition, dynamic field classification, embedded multi-round retrieval, a comparative learning mechanism and a generation optimization strategy. According to the method, the query intention can be accurately captured, the most relevant knowledge base can be dynamically selected, the high-precision traceability result is generated through multi-level verification and optimization, the accuracy, correlation and added value of retrieval are remarkably improved, and the high-precision scene requirements of complex policy traceability, compliance check and the like are met.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, information retrieval and generation technology, and more specifically to a policy tracing dynamic retrieval method based on multi-level intent recognition and comparative learning. Background Art

[0002] With the rapid growth in the number of policy documents and regulatory texts and the significant increase in the complexity and diversity of their content, how to efficiently and accurately trace policies and retrieve information has become an important technical challenge.

[0003] Existing retrieval methods mainly rely on keyword matching or rule-based retrieval models. These methods often have difficulty accurately understanding the user's intention when dealing with complex queries, resulting in insufficient relevance and accuracy of retrieval results. At the same time, when dealing with large-scale and complex policy texts, existing methods lack the ability to deeply understand semantics, dynamically optimize, and analyze contextual associations. The traceability results generated are relatively simple and lack substantial added value.

[0004] Therefore, there is an urgent need for a method that integrates multi-level intent recognition, comparative learning, and generative optimization to achieve accurate policy traceability and dynamic retrieval optimization to meet the needs of high-precision application scenarios such as complex policy review and regulatory tracing. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the existing retrieval methods mainly rely on keyword matching or rule-based retrieval models, resulting in insufficient relevance and accuracy of retrieval results.

[0006] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0007] The policy traceability dynamic retrieval method based on multi-level intention recognition and contrastive learning includes the following steps:

[0008] S1, receiving and standardizing the query content input by the user to generate preprocessed text data;

[0009] S2. Use the pre-trained semantic embedding model to embed the pre-processed text data, convert the text data into a high-dimensional semantic vector representation, generate an embedding vector and store it in the knowledge base;

[0010] S3. Through multi-level intent recognition, the query content entered by the user is deeply analyzed to extract the core needs, and the knowledge base is dynamically selected based on the field classification to adjust the search scope of the knowledge base according to different core needs;

[0011] S4. In the knowledge base, multiple rounds of dynamic retrieval driven by embedded vectors and fuzzy semantic expansion matching are used to retrieve the query content input by the user to obtain the retrieval content;

[0012] S5. Optimize the search content through multi-layer verification and comparative learning mechanism to obtain optimized search content;

[0013] S6, using a multi-level attention mechanism combined with paragraph focusing technology to conduct in-depth analysis of the optimized search content and extract the core content most relevant to the query content entered by the user;

[0014] S7. Generate and optimize the core content of paragraph focus and extraction, and generate policy tracing results through multiple rounds of dynamic adjustment strategies;

[0015] S8. Display the generated policy tracing results to the user.

[0016] Furthermore, the query content input by the user in step S1 includes but is not limited to text data of policy documents, legal provisions, and management points.

[0017] Furthermore, the standardization preprocessing in step S1 includes: word segmentation, stop word removal, syntactic analysis, synonym replacement, capitalization standardization, and punctuation cleaning.

[0018] Furthermore, the semantic embedding model in step S2 includes but is not limited to BERT and RoBERTa.

[0019] Furthermore, the step S3 mainly includes the following steps:

[0020] S31, multi-level intent recognition, using pre-trained semantic embedding models to semantically vectorize user queries and convert query content into high-dimensional semantic representations; hierarchical clustering is used to perform hierarchical clustering analysis on these semantic vectors, and vectors with similar semantic features are classified into the same level; on this basis, semantic grouping strategies are applied to refine and optimize the hierarchical groups; finally, a reasonable priority is given to the intent features of each level through an adaptive weight adjustment mechanism, so as to extract the core intent of the user query in a hierarchical manner;

[0021] S32. Combined with the results of multi-level intent recognition, a multi-label classification model is used to accurately match the user's query content to the most relevant field. Based on the classification results, the corresponding field knowledge base is automatically selected for subsequent retrieval and content matching.

[0022] Furthermore, the step S4 mainly includes the following steps:

[0023] S41, multiple rounds of dynamic retrieval, based on the embedded vector generated in step S2, perform preliminary retrieval, and extract the retrieval content that is closest to the query content input by the user from the knowledge base obtained in step S3 by calculating the similarity of the embedded vector;

[0024] S42, fuzzy semantic expansion matching, which captures the content that is semantically similar to the query content entered by the user but expressed differently through fuzzy matching, and combines semantic expansion with synonym expansion and semantic network to optimize the search conditions;

[0025] S43. After the initial search, enter the multi-round dynamic optimization stage to gradually adjust and optimize the search conditions. Each round of search will dynamically adjust the search parameters based on the feedback of the previous round of results to expand the search coverage and obtain the final search content.

[0026] Furthermore, the step S5 mainly includes the following steps:

[0027] S51, after completing the preliminary search, perform multi-layer verification on the search content obtained in step S4, and evaluate and screen the search results by calculating the similarity between the search content and the query content input by the user;

[0028] S52. Construct a contrastive learning model to distinguish the relevant and irrelevant parts of the search content. For the results that are semantically similar but actually irrelevant, use the contrastive learning mechanism to optimize to improve the screening accuracy.

[0029] S53. For the irrelevant parts, a feedback loop mechanism is started, and the verification result will be directly fed back to step S4 to adjust the search content and perform new search content iteration.

[0030] Furthermore, the step S6 mainly includes the following steps:

[0031] S61, performing paragraph focusing on the optimized search content obtained in step S5, calculating the semantic similarity between the paragraph and the query, analyzing the semantics and structure of each paragraph, giving priority to paragraphs that are highly relevant to the user query, and identifying the most valuable text fragments;

[0032] S62. Introduce a multi-level Attention mechanism. Based on paragraph focus, the Attention mechanism allocates attention to text fragments at different levels to identify high-attention information.

[0033] S63. Extract high-interest information to obtain core content that is most relevant to the query content input by the user.

[0034] Furthermore, the step S7 mainly includes the following steps:

[0035] S71, focusing on the paragraphs and extracting the core content most relevant to the query content input by the user, and continuously optimizing the generation process through multiple rounds of dynamic adjustment strategies;

[0036] S72. Use content fusion to organically combine the search content with the generated content, and screen and integrate the policy tracing results through semantic analysis and context relevance evaluation;

[0037] S73. Through multiple rounds of optimization and content fusion, high-quality policy tracing results are finally generated;

[0038] S74. Conduct quality assessment on the generated high-quality policy traceability results, and optimize the generation strategy based on user feedback. The assessment process includes semantic similarity detection, content redundancy check, and logical consistency analysis to ensure the high standards of the final output;

[0039] S75. During the generation process, the generated policy tracing results are verified again for relevance to the query content initially input by the user, and if necessary, they are fed back to step S71 to re-optimize the generation conditions.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention solves the problem of insufficient understanding of complex policy texts and poor relevance of retrieval results in existing policy tracing and retrieval methods by integrating multi-level intent recognition, comparative learning and generation optimization methods.

[0042] 2. The present invention uses multi-level intent recognition technology to accurately analyze the core needs of the query, and combined with the dynamic field classification mechanism, it can automatically select the most relevant knowledge base, reduce interference from irrelevant information, and improve the pertinence and accuracy of the retrieval.

[0043] 3. The present invention adopts embedded multi-round search strategy and fuzzy semantic expansion technology to dynamically adjust search conditions, expand search coverage, and ensure comprehensive coverage of query-related content. Combined with comparative learning mechanism, multi-layer verification of search results is performed to exclude semantically similar but irrelevant content and optimize result quality.

[0044] 4. Through the combination of multi-level Attention mechanism and paragraph focusing technology, the present invention can accurately identify and extract core content that is highly relevant to the query in complex texts, avoid information redundancy and omission of key information, and ensure that the generated results are highly relevant and valuable.

[0045] 5. The present invention uses a dynamically adjusted generation optimization mechanism to perform multiple rounds of dynamic adjustments on the extracted core content, and organically integrates the search content with the newly generated content, avoiding simple information piling, providing more coherent and value-added tracing results, and improving output quality.

[0046] 6. The present invention continuously adjusts and optimizes the processing flow by displaying the traceability results and collecting user feedback. User feedback is directly used to improve the method, forming a closed-loop optimization that can dynamically adapt to the query needs of different users and continuously improve performance and output accuracy.

[0047] 3. The present invention can be effectively applied in various complex scenarios such as policy tracing and compliance checking. It is particularly suitable for retrieval needs that require high accuracy and relevance, and provides strong support for policy formulation, compliance assessment and other tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is the overall flow chart of the present invention;

[0049] Figure 2 is a flow chart of step S5 in the present invention;

[0050] Figure 3 is a flow chart of step S6 in the present invention;

[0051] Figure 4 is a flow chart of step S7 in the present invention;

[0052] Figure 5 Schematic diagram of the policy tracing dynamic retrieval method for multi-level intent recognition and contrastive learning in the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-4 , a policy traceability dynamic retrieval method based on multi-level intention recognition and contrastive learning, specifically including the following steps:

[0055] S1, receiving and standardizing the query content input by the user to generate preprocessed text data;

[0056] The query content input by the user is received through the front-end interface. The content input by the user may contain complex text data such as policy documents, legal provisions, and management points. To ensure the effectiveness and accuracy of subsequent processing, the input text data is pre-processed in a standardized manner, including the following steps:

[0057] S11, data reception:

[0058] The front end receives the query content input by the user and collects the input text data. This data may be unstructured policy documents, regulatory content, or management key points, and the data is complex and diverse.

[0059] S12. Word Segmentation:

[0060] Use natural language processing (NLP) technology to perform word segmentation on the received text, splitting the text into independent lexical units for subsequent analysis and processing.

[0061] S13. Stop Word Removal:

[0062] Remove common stop words that have no actual semantic contribution (such as "de", "le", "zai", etc.), reduce text noise, and retain keyword vocabulary that contributes to semantic understanding.

[0063] S14. Syntactic Analysis:

[0064] Perform syntactic analysis to understand the structure of the text, identify the relationships between words, enhance the depth of understanding of the text content, and provide structured syntactic information for subsequent semantic embedding generation.

[0065] S15. Data Normalization:

[0066] Perform normalization processing on the input text, such as unifying case, cleaning punctuation marks, and replacing synonyms, etc., to improve the consistency and standardization of text processing.

[0067] S16. Preprocessing Output:

[0068] The preprocessed text data is converted into a more normalized and structured input form, preparing for subsequent semantic embedding generation and efficient retrieval.

[0069] Through the above steps, it is possible to effectively perform standardized preprocessing on the user input query, reduce noise and redundant information in the text, and provide a solid foundation for in-depth semantic understanding and subsequent retrieval of policy texts.

[0070] S2. Use a pre-trained semantic embedding model to perform embedding generation on the preprocessed text data, convert the text data into a high-dimensional semantic vector representation, generate embedding vectors and store them in the knowledge base. The semantic embedding model includes but is not limited to BERT, RoBERTa;

[0071] Use the pre-trained semantic embedding model BERT to perform embedding generation on the preprocessed text data, converting the text content into a high-dimensional semantic vector representation. Through the embedding model, capture the deep semantic features of the text, including context relationships, dependencies between words, and implicit semantic associations, ensuring that each policy document, regulatory clause, or user query is accurately mapped to the high-dimensional vector space.

[0072] The generated high-dimensional semantic vector retains the core semantic information of the text, making similar text content closer in the vector space, which is convenient for subsequent rapid retrieval and comparison. The embedding vector supports a deep understanding of the text content and provides a unified semantic representation for different types of text, enhancing the method's ability to handle diverse inputs.

[0073] The generated embedding vectors are efficiently stored in the knowledge base, combined with the corresponding field classification and metadata information, to provide a data basis for subsequent dynamic retrieval, comparative learning and content generation optimization. Vectorized storage can quickly perform similarity calculations and content matching, improve retrieval efficiency and response speed, and ensure that accurate and relevant traceability results can be provided when processing complex policy queries.

[0074] S3. Through multi-level intent recognition, the query content entered by the user is deeply analyzed to extract the core needs, and the knowledge base is dynamically selected based on the field classification to adjust the search scope of the knowledge base according to different core needs;

[0075] Through multi-level intent recognition technology, we deeply analyze the query content entered by the user, accurately extract the core needs, and realize dynamic knowledge base selection in combination with field classification. This process can accurately understand the user's query intention and conduct subsequent retrieval in a targeted manner.

[0076] S31, Multi-level Intent Recognition:

[0077] The pre-trained semantic embedding model is used to semantically vectorize user queries and convert the query content into a high-dimensional semantic representation. Hierarchical clustering is used to perform hierarchical clustering analysis on these semantic vectors and classify vectors with similar semantic features into the same level. On this basis, the semantic grouping strategy is applied to refine and optimize the hierarchical groups. Finally, an adaptive weight adjustment mechanism is used to assign reasonable priorities to the intent features of each level, thereby extracting the core intent of the user query in a hierarchical manner.

[0078] S32. Field classification and dynamic knowledge base selection:

[0079] Combined with the results of multi-level intent recognition, a multi-label classification model is used to accurately match the user's query content to the most relevant fields (such as finance, agriculture, justice, science and education, etc.). This multi-label classification can effectively distinguish the dominant intent in complex multi-intent queries and set an accurate field scope for subsequent retrieval. According to the classification results, the corresponding field knowledge base is automatically selected for subsequent retrieval and content matching. The dynamic field knowledge base selection mechanism can automatically adjust the search scope according to different intents, reduce interference from irrelevant fields, and thus improve the efficiency and accuracy of retrieval.

[0080] S4. In the knowledge base, multiple rounds of dynamic retrieval driven by embedded vectors and fuzzy semantic expansion matching are used to retrieve the query content input by the user to obtain the retrieval content;

[0081] An embedded vector-driven multi-round dynamic retrieval strategy is adopted, combined with fuzzy semantic expansion technology, to ensure comprehensive retrieval of relevant policy content in user queries, thereby improving retrieval coverage and accuracy.

[0082] S41, embedded multi-round search strategy:

[0083] First, the Dense Passage Retrieval (DPR) technique is used to perform a preliminary search based on the embedding vector generated in step S2. In this stage, the search content that is closest to the user query is extracted from the knowledge base through similarity calculation of the embedding vector.

[0084] S42, fuzzy matching and semantic extension:

[0085] In the process of multiple rounds of retrieval, in order to make up for the shortcomings of the initial search, a fuzzy matching strategy is introduced. Through fuzzy matching, it is possible to capture content that is semantically similar to the user's query but expressed differently, avoiding the omission of relevant policies due to the limitation of exact matching. At the same time, combined with semantic expansion technology, synonym expansion and semantic networks (such as WordNet) are used to enrich the search conditions. Semantic expansion can identify synonyms, near-synonyms and related words, expand the search scope, and ensure that all potentially relevant policy texts are fully covered.

[0086] S43. After the initial search, enter the multi-round dynamic optimization stage to gradually adjust and optimize the search conditions:

[0087] After the initial search, multiple rounds of dynamic optimization are entered to gradually adjust and optimize the query conditions. Each round of search will dynamically adjust the search parameters (such as similarity threshold, query expansion range, etc.) based on the feedback of the previous round of results, expand the coverage of the search, ensure that all possible relevant content is retrieved, and obtain the final search content.

[0088] S44, Dynamic Adjustment and Iterative Optimization:

[0089] The combination of embedded multi-round retrieval strategy and fuzzy semantic expansion can perform dynamic multi-round iterative retrieval in the embedded space and gradually optimize the retrieval content. Each round of retrieval depends on the feedback of the previous round. Through continuous adjustment of retrieval conditions and semantic expansion, the most relevant policy content is gradually approached. This method not only improves the coverage of retrieval, but also optimizes the matching degree of content in complex query scenarios, and can respond to diverse user query needs more intelligently.

[0090] By combining multiple rounds of dynamic retrieval and fuzzy semantic expansion, this step achieves comprehensive coverage and accurate retrieval of policy content, providing users with more relevant and complete traceability results.

[0091] S5. Optimize the search content through multi-layer verification and comparative learning mechanism to obtain optimized search content;

[0092] The retrieved content is optimized through multi-layer verification and comparative learning mechanisms to ensure the relevance and accuracy of the final output results. This process continuously improves the quality of the results through careful screening and feedback loops of the search results.

[0093] S51. Content verification and optimization:

[0094] After the initial search is completed, the search content obtained in step S4 is subjected to multiple layers of verification to ensure that the search content is highly relevant to the query content entered by the user. Content verification is not limited to surface similarity, but also includes in-depth analysis at the semantic level. By calculating the semantic similarity between the content and the query, the search results are evaluated and screened to exclude irrelevant or misdetected content.

[0095] S52. Contrastive learning driven optimization process:

[0096] For results that are semantically similar but actually irrelevant, we use the contrastive learning mechanism to optimize and improve screening accuracy. The contrastive learning mechanism plays a key role in the optimization process. By building a contrastive learning model, we can effectively distinguish between relevant and irrelevant parts of the search content, especially when facing semantically similar interference items, and accurately exclude irrelevant content. Through continuous iterative training and adjustment of search content, contrastive learning can continuously improve recognition capabilities in complex semantic scenarios, ensuring that the final output results are both highly relevant and accurately reflect the user's query intent.

[0097] S53. Feedback loop and search strategy adjustment:

[0098] If the results are found to deviate from the user's query intent during content verification and comparative learning, a feedback loop mechanism will be initiated. The verification results will be directly fed back to the embedded multi-round retrieval and fuzzy matching in step S4 to adjust the retrieval conditions and perform new retrieval iterations. This closed-loop feedback mechanism can dynamically adjust the retrieval strategy, continuously optimize the accuracy and relevance of the results, achieve self-learning and adaptation, and thus enhance overall intelligence.

[0099] S54, content verification and retrieval feedback integration:

[0100] The verification results are not only used to screen and optimize the current search content, but also to adjust the subsequent search strategy. The feedback from content verification is directly integrated into the search strategy adjustment process, so that the search and optimization processes complement each other and form an adaptive closed-loop optimization. This mechanism can maintain high efficiency and accuracy in complex policy text environments, ensuring that users can obtain the most relevant and optimized search content when facing multi-intention and multi-field queries.

[0101] Through the combination of multi-layer verification and comparative learning, the search results can be deeply screened and optimized, the relevance and accuracy of the content can be continuously improved, and ultimately users can be provided with high-quality policy traceability results that meet the query intent.

[0102] S6, using a multi-level attention mechanism combined with paragraph focusing technology to conduct in-depth analysis of the optimized search content and extract the core content most relevant to the query content entered by the user;

[0103] The multi-level Attention mechanism is combined with paragraph focusing technology to conduct in-depth analysis of the search content and accurately extract the core content that is most relevant to the query content entered by the user. This process ensures that the output results can accurately reflect the user's query intention by performing multi-level processing on the semantics and structure of the paragraph.

[0104] S61, paragraph focus technology:

[0105] First, the retrieved text results are paragraph-focused, the semantics and structure of each paragraph are analyzed, and the paragraphs that are highly relevant to the query entered by the user are prioritized. Paragraph focusing identifies the most valuable text fragments by calculating the semantic similarity between the paragraph and the query.

[0106] During the focusing process, not only the matching of keywords is considered, but also the contextual association and logical structure of the paragraph are deeply analyzed to ensure that the extracted paragraph can fully reflect the core theme of the user's query.

[0107] S62, Multi-level Attention Mechanism:

[0108] In order to accurately identify the core content of a paragraph, a multi-level Attention mechanism is introduced. Based on paragraph focus, the Attention mechanism allocates different levels of attention to sentences and words in a paragraph to identify high-attention information.

[0109] Multi-level attention allows deep focus at the sentence level, word level, and semantic level, improving the ability to extract relevant information by strengthening attention to key sentences and words. This mechanism effectively enhances the understanding of complex texts and avoids missing key details.

[0110] S63, core content extraction:

[0111] After paragraph focusing and multi-level attention processing, the core content of high-attention information is extracted. The core content extraction aims to mine the text fragments that are most closely related to the user's query and ensure that the generated results accurately reflect the user's query intention.

[0112] Combining the context semantics and Attention allocation results, the content with high relevance and information value in the paragraph is extracted, avoiding information redundancy and interference from irrelevant content.

[0113] S64, Intent-based Precision Positioning:

[0114] The combination of paragraph focusing and multi-level attention mechanisms can accurately locate relevant text fragments in complex policy documents based on user intent. This intent-based positioning method significantly improves the ability to analyze text and the accuracy of content extraction, ensuring the high relevance of tracing results.

[0115] Through the dual alignment of semantics and intent, paragraph focus is no longer limited to surface similarities, but a deep understanding of the intrinsic connections of the text to achieve accurate responses to user queries.

[0116] Through the paragraph focusing and core content extraction of the multi-level Attention mechanism, this step effectively improves the ability to deeply analyze the text and extract relevant information, providing users with more accurate policy tracing results and enhancing the response effect in complex query scenarios.

[0117] S7. Generate and optimize the core content of paragraph focus and extraction, and generate policy tracing results through multiple rounds of dynamic adjustment strategies;

[0118] The results of paragraph focusing and core content extraction are generated and optimized, and high-quality policy tracing results are generated through multiple rounds of dynamic adjustment strategies. The content fusion method is used to organically combine the retrieved content with the generated new content to improve the accuracy and logical coherence of the output results.

[0119] S71. Generation optimization and dynamic adjustment:

[0120] First, the text obtained from the paragraph focusing and core content extraction stages is initially generated, and the generation process is continuously optimized through multiple rounds of dynamic adjustment strategies. The generation optimization mechanism combines context understanding, semantic consistency checking, and logical structure adjustment to make the generated content more consistent with the user's query intent. Dynamic adjustment involves repeated iterations of the generated content, gradually correcting the generation strategy by evaluating the relevance, semantic integrity, and logical coherence of the content. Each round of adjustment is refined based on the output of the previous round, making the final result more coherent and accurate.

[0121] S72. Content integration strategy:

[0122] The innovative content fusion strategy is adopted to organically combine the search content with the generated content. The fusion strategy is not to simply splice the two parts of content, but to screen and integrate the most valuable information through semantic analysis and contextual relevance evaluation. In the process of content fusion, the search content is reorganized through semantic matching and logical association, so that the generated content is naturally connected with the existing text, avoiding duplication and piling of information. This process not only improves the overall quality of the content, but also enhances the added value of the output results.

[0123] S73, high quality output generation:

[0124] Through multiple rounds of optimization and content fusion, high-quality policy tracing results are finally generated. The output content not only accurately reflects the core intent of the user's query, but also has a certain depth of analysis and comprehensive information, providing users with value beyond the search content itself. In the generation process, the policy background, relevant regulations and contextual information are comprehensively considered to ensure that the output results are logically clear and semantically consistent, so that users can more intuitively understand the key points of policy tracing.

[0125] S74, Output quality assessment and feedback optimization:

[0126] The quality of the generated final output is evaluated, and the generation strategy is optimized in combination with user feedback. The evaluation process includes semantic similarity detection, content redundancy check, and logical consistency analysis to ensure the high standard of the final output. The feedback mechanism allows the generation model and fusion strategy to be continuously improved based on the user's evaluation of the generated content, achieving adaptive content generation optimization. In this way, the output quality can be continuously improved when facing diverse queries.

[0127] S75, Feedback Optimization Mechanism:

[0128] During the generation process, the relevance of the generated content to the query content initially input by the user will be verified again, and if necessary, it will be fed back to step S71 to re-optimize the generation conditions to ensure that the final output content is logically coherent and semantically clear.

[0129] S8, display feedback of traceability results and overall improvement based on user opinions;

[0130] The generated policy tracing results are presented to users, and user feedback is collected to continuously improve performance.

[0131] like Figure 5 As shown, the present invention also proposes a policy traceability dynamic retrieval process summary based on multi-level intent recognition and contrastive learning, including:

[0132] The user query processing process 410 is used to receive the query content input by the user and perform standardized preprocessing on the input query content, including operations such as word segmentation, removal of stop words, and syntactic analysis, so as to generate standardized text data suitable for subsequent semantic processing.

[0133] The semantic embedding generation process 420 is used to use the pre-trained model to perform high-dimensional semantic embedding on the standardized text data, generate high-dimensional vectors that reflect the deep semantics of the text, and store these vectors in the knowledge base to provide basic data support for subsequent retrieval and comparative learning.

[0134] The multi-level intent recognition and classification process 430 is used to perform multi-level intent recognition on the query content input by the user, combine semantic embedding, hierarchical clustering and adaptive weight adjustment and other technologies to accurately extract the core intent of the user query, and map the recognition results to the knowledge base of the corresponding field for classification.

[0135] The embedding vector retrieval and optimization process 440 is used to perform multiple rounds of dynamic retrieval in the embedding space, optimize the preliminary retrieval results through the DensePassage Retrieval technology, and dynamically adjust the retrieval conditions in combination with the fuzzy matching strategy and semantic expansion technology to expand the retrieval scope and improve the accuracy of the retrieval results.

[0136] The content verification and comparative learning process 450 is used to perform multi-level verification on the initially retrieved content, optimize the search results through a comparative learning mechanism, exclude semantically similar but irrelevant content, and dynamically adjust the search strategy based on the verification feedback to continuously optimize the relevance of the output results.

[0137] The paragraph focusing and core extraction process 460 is used to perform paragraph focusing analysis on the retrieved text through a multi-level Attention mechanism, identify the most relevant paragraphs, and extract the core content to ensure that the generated tracing results can accurately reflect the user's query intention.

[0138] The generation optimization and content fusion process 470 is used to perform multiple rounds of dynamic adjustments on the results of paragraph focusing and core content extraction, combine the retrieved content with the newly generated content through the generation optimization mechanism, implement the content fusion strategy, and generate coherent and value-added policy tracing results.

[0139] The result display and feedback process 480 is used to display the generated policy tracing results to the user and collect user feedback information. The processing processes are adjusted and optimized according to user feedback to continuously improve the overall performance and accuracy of the output results.

[0140] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. The patent protection scope of the present invention shall be based on the claims. All equivalent structural changes made using the contents of the description and drawings of the present invention should also be included in the protection scope of the present invention.

Claims

1. A policy traceability dynamic retrieval method based on multi-level intention recognition and contrastive learning, characterized by: The following steps are involved: S1, receiving and standardizing the query content input by the user to generate preprocessed text data; S2. Use the pre-trained semantic embedding model to embed the pre-processed text data, convert the text data into a high-dimensional semantic vector representation, generate an embedding vector and store it in the knowledge base; S3. Through multi-level intent recognition, the query content entered by the user is deeply analyzed to extract the core needs, and the knowledge base is dynamically selected based on the field classification to adjust the search scope of the knowledge base according to different core needs; S4. In the knowledge base, multiple rounds of dynamic retrieval driven by embedded vectors and fuzzy semantic expansion matching are used to retrieve the query content input by the user to obtain the retrieval content; S5. Optimize the search content through multi-layer verification and comparative learning mechanism to obtain optimized search content; S6. Use a multi-level attention mechanism combined with paragraph focusing technology to conduct in-depth analysis of the optimized search content and extract the core content most relevant to the query content entered by the user; S7. Generate and optimize the core content of paragraph focus and extraction, and generate policy tracing results through multiple rounds of dynamic adjustment strategies; S8. Display the generated policy tracing results to the user.

2. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The query content input by the user in step S1 includes but is not limited to text data of policy documents, legal provisions, and management points.

3. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The standardization preprocessing in step S1 includes: word segmentation, stop word removal, syntactic analysis, synonym replacement, capitalization standardization, and punctuation cleaning.

4. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The semantic embedding model in step S2 includes but is not limited to BERT and RoBERTa.

5. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The step S3 mainly includes the following steps: S31, multi-level intent recognition, using pre-trained semantic embedding models to semantically vectorize user queries and convert query content into high-dimensional semantic representations; hierarchical clustering is used to perform hierarchical clustering analysis on these semantic vectors, and vectors with similar semantic features are classified into the same level; on this basis, semantic grouping strategies are applied to refine and optimize the hierarchical groups; finally, a reasonable priority is given to the intent features of each level through an adaptive weight adjustment mechanism, so as to extract the core intent of the user query in a hierarchical manner; S32. Combined with the results of multi-level intent recognition, a multi-label classification model is used to accurately match the user's query content to the most relevant field. Based on the classification results, the corresponding field knowledge base is automatically selected for subsequent retrieval and content matching.

6. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The step S4 mainly includes the following steps: S41, multiple rounds of dynamic retrieval, based on the embedded vector generated in step S2, perform preliminary retrieval, and extract the retrieval content that is closest to the query content input by the user from the knowledge base obtained in step S3 by calculating the similarity of the embedded vector; S42, fuzzy semantic expansion matching, which captures the content that is semantically similar to the query content entered by the user but expressed differently through fuzzy matching, and combines semantic expansion with synonym expansion and semantic network to optimize the search conditions; S43. After the initial search, enter the multi-round dynamic optimization stage to gradually adjust and optimize the search conditions. Each round of search will dynamically adjust the search parameters based on the feedback of the previous round of results to expand the search coverage and obtain the final search content.

7. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The step S5 mainly includes the following steps: S51, after completing the preliminary search, perform multi-layer verification on the search content obtained in step S4, and evaluate and screen the search results by calculating the similarity between the search content and the query content input by the user; S52. Construct a contrastive learning model to distinguish the relevant and irrelevant parts of the search content. For the results that are semantically similar but actually irrelevant, use the contrastive learning mechanism to optimize to improve the screening accuracy. S53. For the irrelevant parts, a feedback loop mechanism is started, and the verification result will be directly fed back to step S4 to adjust the search content and perform new search content iteration.

8. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The step S6 mainly includes the following steps: S61, performing paragraph focusing on the optimized search content obtained in step S5, calculating the semantic similarity between the paragraph and the query, analyzing the semantics and structure of each paragraph, giving priority to paragraphs that are highly relevant to the user query, and identifying the most valuable text fragments; S62. Introduce a multi-level Attention mechanism. Based on paragraph focus, the Attention mechanism allocates attention to text fragments at different levels to identify high-attention information. S63. Extract high-interest information to obtain core content that is most relevant to the query content input by the user.

9. The policy source tracing dynamic retrieval method based on multi-level intention recognition and contrastive learning according to claim 1 is characterized by: The step S7 mainly includes the following steps: S71, focusing on the paragraphs and extracting the core content most relevant to the query content input by the user, and continuously optimizing the generation process through multiple rounds of dynamic adjustment strategies; S72. Use content fusion to organically combine the search content with the generated content, and screen and integrate the policy tracing results through semantic analysis and context relevance evaluation; S73. Through multiple rounds of optimization and content fusion, high-quality policy tracing results are finally generated; S74. Conduct quality assessment on the generated high-quality policy traceability results, and optimize the generation strategy based on user feedback. The assessment process includes semantic similarity detection, content redundancy check, and logical consistency analysis to ensure the high standards of the final output; S75. During the generation process, the generated policy tracing results are used to verify the relevance to the query content initially input by the user, and can be fed back to step S71 to re-optimize the generation conditions.

Images