Large-model fine-grained illusion detection and correction method and product

By constructing entity sets and calculating support scores to locate suspicious entities, and combining this with minimal correction using a large language model, the illusion problem in content generated by the large language model is solved, improving the accuracy and efficiency of detection and correction.

CN121996997APending Publication Date: 2026-05-08MILITARY SCI INFORMATION RES CENT ACAD OF MILITARY SCI OF THE CHINESE PEOPLES LIBERATION ARMY
0 Cites 0 Cited by

Patent Information

Application Number
CN202511999991.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-29
Publication Date
2026-05-08

AI Technical Summary

Technical Problem

Existing large language models are prone to illusions when generating content, and existing detection methods rely solely on information element matching or require complex internal analysis, resulting in high correction costs and difficulty in accurately locating entities at the entity level.

Method used

By constructing entity sets of reference information and test content, support scores are calculated to locate suspicious entities, and a large language model is used for minimal correction to avoid in-depth analysis within the model.

Benefits of technology

It significantly improves factual consistency and error correction accuracy without changing the original sentence structure and writing logic, while reducing engineering complexity and resource consumption and maintaining text stability.

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

Abstract

The invention discloses a large-model fine-grained illusion detection and correction method and product. The method comprises the following steps: constructing reference information according to titles and description information of reference literatures, extracting entities in the reference information through a large language model, obtaining coded representation of each entity, and constructing a reference information entity set; entities in the to-be-tested content are extracted through the large language model, the coded representation of each entity is obtained, and a to-be-tested content entity set is constructed; for each entity in the to-be-tested content entity set, respectively calculating a support degree score between the entity and the entity in the reference information entity set, determining whether the entity in the to-be-tested content entity set is supported by reference literatures according to the support degree scores, and positioning suspicious entities; and on the basis of the positioned suspicious entity, constructing a hallusion correction cue word, judging whether the suspicious entity relates to the hallusion content or not by utilizing the text understanding capability of the large language model, performing minimum amplitude correction on the involved hallusion content, and outputting the corrected content.
Need to check novelty before this filing date? Find Prior Art