Enhanced retrieval generation optimization method for private domain knowledge base

Abstract

The invention provides an enhanced retrieval generation optimization method for a private domain knowledge base, and belongs to the technical field of knowledge base construction.The method comprises the steps that a labeling mechanism is introduced, labels are added into document data, and a knowledge retrieval range can be defined according to user information before retrieval; a mixed retrieval mode combining vector retrieval and full-text retrieval is adopted; and for multiple result sets obtained in different retrieval modes, adopting a result reordering technology of fusing reciprocal ordering with RRF to screen out TOP-K documents most related to user query. The illusion phenomenon possibly generated by the model can be effectively reduced, and the accuracy and reliability of a retrieval result are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge base construction, user access control, enhanced retrieval generation, result rearrangement, etc., and in particular to an enhanced retrieval generation optimization method for a private domain knowledge base. BACKGROUND

[0002] Traditional knowledge base management often lacks fine-grained permission management for knowledge and cannot achieve effective user access control. Existing retrieval generation methods are mostly based on vector retrieval, which can efficiently obtain semantically related results, but lack support for precise text matching. Even if the information precision matching capability is improved, it still cannot fully meet the user's query needs. SUMMARY

[0003] To solve the above technical problems, the present application provides an enhanced retrieval generation optimization method for a private domain knowledge base, which performs access control on knowledge base query maintenance to ensure the security of the knowledge base. The enhanced retrieval generation process is optimized through hybrid query and result rearrangement to improve the comprehensiveness and accuracy of the results.

[0004] The technical solution of the present application is:

[0005] An enhanced retrieval generation optimization method for a private domain knowledge base introduces a labeling mechanism, adds labels to document data, and can define the knowledge retrieval range according to user information before retrieval.

[0006] A hybrid retrieval method combining vector retrieval and full-text retrieval is used.

[0007] For multiple result sets obtained by different retrieval methods, a result reordering technology using reverse ranking fusion RRF is used to filter out the TOP-K documents most relevant to the user query.

[0008] Further,

[0009] Different users' needs and responsibilities for the knowledge base are analyzed, and corresponding roles are defined according to different functional needs. A role-based permission management is established to perform fine-grained permission control on different documents in the knowledge base, ensuring that users can only access authorized content.

[0010] Further,

[0011] A hybrid retrieval method is used to process user queries to obtain a comprehensive hybrid result set.

[0012] A text embedding model is used to convert the user query into a dense vector, and TOP-K results are found in the vector database according to the vector similarity as the vector retrieval result set.

[0013] The word segmenter is used to obtain the query keywords from the user's query, and the full-text search engine is used to match the keywords to obtain the matching full-text search results.

[0014] Furthermore,

[0015] Improve the relevance of query results by evaluating and reordering search results.

[0016] By aggregating multiple result sets from vector retrieval and full-text retrieval, and using a reciprocal ranking fusion algorithm, the ranking scores of items in the result sets are recalculated.

[0017] The result set is reordered based on the ranking, and the TOP-K results are output.

[0018] The beneficial effects of this invention are

[0019] Through optimization methods such as access control, hybrid query retrieval, and result rearrangement and merging, the enhanced retrieval results generated for the private domain knowledge base have been significantly improved. This not only enhances the security and data privacy protection capabilities of the knowledge base but also strengthens the flexibility and accuracy of retrieval, providing users with a more efficient and precise search service. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. 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.

[0021] Traditional knowledge base management often lacks granular access control, failing to achieve effective user access control. To enhance data security, a tagging mechanism is introduced, adding tags to document data to accurately define the knowledge retrieval scope based on user information before searching. Meanwhile, given that most existing retrieval generation methods are based on vector retrieval, while efficiently obtaining semantically relevant results, they lack support for precise text matching. Therefore, a hybrid retrieval method combining vector retrieval and full-text retrieval is proposed. This improves the accuracy of information matching and more comprehensively meets users' query needs. To further optimize retrieval results, a reverse ranking fusion (RRF) technique is used to re-rank the results from multiple result sets obtained by different retrieval methods, selecting the top-K documents most relevant to the user's query. This effectively reduces the "illusion" phenomenon that the model may produce, improving the accuracy and reliability of retrieval results.

[0022] With the rise of large language models, traditional knowledge retrieval, represented by retrieval-enhanced generation, has brought about revolutionary changes. To prevent unauthorized users from maliciously altering the knowledge base and leaking sensitive or critical information, access controls need to be designed and implemented during knowledge base creation and retrieval to protect its security and ensure that only authorized users can perform the corresponding operations. To obtain comprehensive and accurate knowledge base retrieval results, multiple retrieval strategies are often employed, and multiple query result sets are merged through result reordering to generate a unified ranking result set, ensuring broad coverage and high relevance of the result set.

[0023] Specifically as follows:

[0024] 1. Access Control

[0025] Analyze the needs and responsibilities of different users for the knowledge base, define corresponding roles according to different functional requirements, establish role-based access control, and implement fine-grained access control for different documents in the knowledge base to ensure that users can only access authorized content.

[0026] 2. Mixed query retrieval

[0027] Multiple query retrieval strategies are employed to process user queries and obtain a comprehensive and mixed result set. A text embedding model is used to transform user queries into dense vectors, and the TOP-K results are found in the vector database based on vector similarity, forming the vector retrieval result set. Simultaneously, a word segmenter is used to obtain query keywords from the user queries, and a full-text search engine is used to match these keywords, obtaining precisely matched full-text search results.

[0028] 3. Result rearrangement and merging

[0029] The relevance of search results is improved by evaluating and reordering them. Multiple result sets, including vector search and full-text search, are aggregated. A reciprocal ranking fusion algorithm is used to recalculate the ranking scores of items in the result sets, reorder the result sets according to the rankings, and output the TOP-K results.

[0030] The knowledge base of this invention has high security and strong data privacy.

[0031] The retrieval method is flexible, the result set has good coverage of the retrieved information, and the recall rate is high.

[0032] The search results are highly relevant and accurate.

[0033] The above description is merely a preferred embodiment of the present invention and is used only to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An enhanced retrieval generation optimization method for a private domain knowledge base, characterized in that, A tagging mechanism is introduced to add tags to document data, and the scope of knowledge retrieval is defined based on user information before retrieval. A hybrid retrieval method combining vector retrieval and full-text retrieval is adopted; For multiple result sets obtained from different retrieval methods, a result re-ranking technique combining inverse sorting and RRF is used to filter out the TOP-K documents most relevant to the user's query.

2. The method according to claim 1, characterized in that, Analyze the needs and responsibilities of different users for the knowledge base, establish role-based access control according to different roles, and implement fine-grained access control for different documents in the knowledge base to ensure that users can only access authorized content.

3. The method according to claim 1, characterized in that, User queries are processed using a hybrid retrieval method to obtain a comprehensive hybrid result set.

4. The method according to claim 3, characterized in that, The text embedding model is used to convert user queries into dense vectors. The top-K results are found in the vector database based on vector similarity and used as the vector retrieval result set.

5. The method according to claim 4, characterized in that, The word segmenter is used to obtain the query keywords from the user's query, and the full-text search engine is used to match the keywords to obtain the matching full-text search results.

6. The method according to claim 1, characterized in that, Improve the relevance of query results by evaluating and reordering search results.

7. The method according to claim 6, characterized in that, By aggregating multiple result sets from vector retrieval and full-text retrieval, and using a reciprocal ranking fusion algorithm, the ranking scores of items in the result sets are recalculated.

8. The method according to claim 7, characterized in that, The result set is reordered based on the ranking, and the TOP-K results are output.