A document-level relation extraction method based on a generative model

By combining a generative model with a lightweight discriminant module, the problem of output uncertainty in generative document-level relation extraction is solved, thereby improving the accuracy and controllability of relation extraction and generating relation summaries that are closer to human intent.

CN119088980BActive Publication Date: 2026-06-02BEIHANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
BEIHANG UNIV
Filing Date
2024-08-23
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing generative document-level relation extraction methods have a large search space and strong randomness in the decision-making process, making it difficult to guarantee the consistency and accuracy of the output. They are also prone to introducing noise and cannot effectively extract the close relationship information between the subject and object in the document.

Method used

We employ a document-level relation extraction method based on generative models, combined with high-precision example guidance and a lightweight discrimination module. The generative model understands document content in complex contexts, generates relation summaries, and maps them to predefined relation categories using the lightweight discrimination module. The nearest neighbor algorithm is introduced to optimize the generation process.

Benefits of technology

It achieves a dual improvement in the accuracy and controllability of relation extraction, generates higher quality and more accurate relation summaries, reduces noise, and improves the model's adaptability and output consistency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088980B_ABST
    Figure CN119088980B_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of information processing, and particularly relates to a document-level relation extraction method based on a generative model, which comprises the following steps: firstly, in the training stage, a generative model is used to compress and refine the document, and the relation summary of the subject and object in the document is extracted, so that information closely related to the target is screened out; secondly, in the testing stage, a lightweight discriminant module is introduced, and the relation summary output by the generative model is mapped to a predefined relation category; and the generative model has self-adaptive learning ability and can dynamically and autonomously learn human carefully selected examples to generate output closer to human intention.
Need to check novelty before this filing date? Find Prior Art