A mask fusion triple extraction method for semantic association in data penetration

By employing a three-stage extraction framework and an adaptive mask attention fusion mechanism, the problems of positive and negative sample imbalance and semantic representation difficulties in triplet extraction methods are solved, achieving high-quality entity recognition and relation extraction, and improving the accuracy and stability of data integration and semantic association.

CN122433740APending Publication Date: 2026-07-21SOUTH CHINA UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTH CHINA UNIV OF TECH
Filing Date
2026-04-23
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing triplet extraction methods face problems of positive and negative sample imbalance and semantic representation difficulties when processing professional domain texts, resulting in insufficient accuracy and robustness of entity recognition and relation extraction, making it difficult to achieve high-quality data integration and semantic association.

Method used

A three-stage extraction framework is adopted, including span filtering, type recognition and relation extraction stages. It combines an adaptive mask attention fusion mechanism and a dual affine interaction mechanism to alleviate the sample imbalance problem and improve the semantic representation capability by adaptively fusing features and enhancing semantic interaction.

Benefits of technology

It significantly improves the accuracy and robustness of entity recognition and relation extraction, generates high-quality semantic triples, and is suitable for data integration and semantic association in complex scenarios, especially showing good application results in industrial chain texts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122433740A_ABST
    Figure CN122433740A_ABST
Patent Text Reader

Abstract

The application discloses a mask fusion triple extraction method for semantic association in data penetration, which is realized based on a three-stage extraction framework. In the span filtering stage, high-quality candidate entities are dynamically screened through binary classification prediction, thereby relieving the imbalance problem of positive and negative samples. In the type identification stage, a mask strategy from the boundary to the center and an adaptive mask attention fusion mechanism are adopted to enhance the candidate entity features and improve the accuracy of entity type identification. In the relationship extraction stage, a double-affine interaction mechanism and an adaptive mask attention fusion mechanism are combined to enhance the semantic interaction between subject and object entities, accurately predict the relationship type, and generate semantic triples. The application effectively relieves the error propagation problem and the imbalance problem of positive and negative samples in the traditional pipeline method, solves the problems of fuzzy entity boundary and difficult semantic representation in the industrial chain text, and provides efficient and reliable technical support for the construction of semantic association in data penetration.
Need to check novelty before this filing date? Find Prior Art