Patent entity relationship identification model training method and device and computer medium

By combining the ELMO model and multi-feature fusion mechanism with a bidirectional tree structure end-to-end relationship classification model, the problems of nested entity recognition and complex sentence structure in patent entity relationship identification are solved, achieving high-precision patent entity relationship recognition and visualization analysis.

CN118133940BActive Publication Date: 2026-06-02CHONGQING UNIV OF POSTS & TELECOMM
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHONGQING UNIV OF POSTS & TELECOMM
Filing Date
2024-03-12
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing patent entity relationship recognition models cannot accurately identify nested entities, and they suffer from information loss and inaccurate relationship extraction when dealing with complex sentence structures.

Method used

An end-to-end relation classification model using the ELMO model combined with part-of-speech information and employing a multi-feature fusion mechanism and a bidirectional tree structure is constructed through entity annotation, relation annotation, and feature extraction. This model includes a representation layer, a decoding layer, a word embedding layer, a sequence layer, and a dependency layer. Pointer networks and a bidirectional tree structure are used for entity and relation recognition.

Benefits of technology

It improves the accuracy and interpretability of patent entity recognition, enabling a better understanding of key information in patent abstracts, generating clear and intuitive recognition results, and reducing the difficulty of patent analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118133940B_ABST
    Figure CN118133940B_ABST
Patent Text Reader

Abstract

The application belongs to the field of patent information processing, and particularly relates to a training method and device of a patent entity relationship identification model and a computer medium. The method is used for training a patent entity relationship identification model, and specifically comprises the following steps: obtaining patent data from a patent database, and preprocessing the patent data to extract abstract texts of the patent data; performing entity annotation and relationship annotation on the abstract texts of the patent data to generate entity training texts and relationship training texts; inputting the entity training texts into a patent entity identification model, and inputting the relationship training texts into a patent entity relationship classification model; and training the patent entity relationship identification model according to the entity annotation, the entity identification result, the relationship annotation and the relationship identification result. The application can improve the effect of the patent entity relationship identification model, and further improve the accuracy of the patent entity relationship identification model.
Need to check novelty before this filing date? Find Prior Art