A retrieval enhancement method and system based on asymmetric locality sensitive hashing

By optimizing the input and output of a large language model using the Asymmetric Locality Sensitive Hash (ALSH) algorithm, a knowledge text database is established and fact checking is performed. This solves the problems of insufficient semantic analysis and low efficiency of hash algorithms in traditional search engines, and achieves an efficient and accurate query process.

CN119046451BActive Publication Date: 2026-07-24HOHAI UNIV
2 Cites 0 Cited by

Patent Information

Application Number
CN202411129574.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2026-07-24
Estimated Expiration
2044-08-16

AI Technical Summary

Technical Problem

Traditional search engines lack semantic analysis and information integration capabilities, resulting in low accuracy and security of keyword searches. Furthermore, existing locality-sensitive hashing algorithms are inefficient in processing non-uniform vector distributions and cannot effectively optimize the output of large language models.

Method used

The asymmetric locality-sensitive hashing (ALSH) algorithm is used to optimize the input and output of a large language model. By establishing a knowledge text database, using fact checking methods to determine text relevance, and storing vectors in a partitioned hash table, the query process is optimized by combining background formulas and hint engineering.

Benefits of technology

It improves the output accuracy and security of large language models, reduces memory requirements and processor load, increases query speed and recall, and solves the problem of low computational efficiency of traditional methods in large databases.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The application discloses a retrieval enhancement method and system based on asymmetric local sensitive hashing. The retrieval enhancement method comprises the following steps: establishing a knowledge text database, including a plurality of knowledge texts; acquiring a query text, and extracting target knowledge texts related to the query text in the knowledge text database based on an asymmetric local sensitive hashing algorithm; establishing a background formula, the background formula including placeholders uniquely corresponding to the target knowledge texts and the query text, and then replacing the corresponding placeholders in the background formula with the target knowledge texts and the query text to obtain an input text; and inputting the input text into a large language model to output an answer text. The application has the characteristics of accuracy and efficiency.
Need to check novelty before this filing date? Find Prior Art