A false entity filtering method based on guideline learning and dual-view gated retrieval

By employing guide learning and dual-view gating retrieval methods, an iteratively updatable guide library and dual-view retrieval module are constructed, solving the problem of unstable deletion of fake entities and achieving high-precision, adaptive fake entity filtering, suitable for scenarios where the upstream model is frozen.

CN122334286APending Publication Date: 2026-07-03JIANGNAN UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
JIANGNAN UNIV
Filing Date
2026-06-03
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies lack sufficient post-processing techniques for freezing upstream models and have insufficient discriminative power for semantic retrieval with few samples, making it difficult to stably delete fake entities and easily leading to false deletions and missed deletions.

Method used

A method based on guide learning and dual-view gating retrieval is adopted. By constructing an iteratively updatable guide library and a dual-view retrieval module, error correction rules are generated. The UCB algorithm mechanism is used to balance the use of guide rules. By combining the dual-view representation of entity identity view and context semantic view, similarity is dynamically weighted and fused to achieve high-precision filtering of fake entities.

Benefits of technology

It achieves high-precision, iterative, and adaptive filtering of fake entities, ensuring filtering quality while also being auditable and interpretable. It is suitable for scenarios where the upstream model is frozen, and no modification to the upstream model is required.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122334286A_ABST
    Figure CN122334286A_ABST
Patent Text Reader

Abstract

This invention relates to the fields of artificial intelligence and natural language processing, and particularly to a method for filtering fake entities based on guide learning and dual-view gating retrieval. The method includes constructing an example library, generating guide rules, and building a guide library; sampling and correcting training prediction results, generating new guide rules for unsuccessful corrections, scoring each guide rule, and updating the guide library; constructing dual views of entities in the test prediction results and entities in the example library, determining the density reliability corresponding to the dual views, obtaining a fusion similarity through gating weights, and obtaining retrieval examples; filtering fake entities in the test prediction results based on the updated guide library and the retrieval examples to obtain post-processed test prediction results. This invention also relates to a fake entity filtering system, apparatus, device, medium, and program product based on guide learning and dual-view gating retrieval.
Need to check novelty before this filing date? Find Prior Art