A dynamic entity prototype based weight network inference method

By optimizing network parameters through dynamic entity prototyping and multi-head attention mechanisms, the problems of high computational complexity, insufficient utilization of semantic information, and low model interpretability in named entity recognition are solved, thereby improving the accuracy and robustness of named entity recognition.

CN116432755BActive Publication Date: 2025-11-28XIDIAN UNIV HANGZHOU RES INST +1
2 Cites 0 Cited by

Patent Information

Application Number
CN202310434601.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-04-21
Publication Date
2025-11-28
Estimated Expiration
2043-04-21

AI Technical Summary

Technical Problem

Existing named entity recognition technologies suffer from high computational complexity, insufficient utilization of semantic information, low model interpretability, and low recognition accuracy due to inaccurate contextual information.

Method used

A weighted network inference method based on dynamic entity prototypes is adopted. By reconstructing the entity recognition dataset, dynamic entity prototypes are generated. Multi-head attention mechanism and BiLSTM/BiGRU network are used for iterative learning to optimize network parameters and improve entity recognition performance.

Benefits of technology

By effectively utilizing local contextual information, the need for additional dictionaries is reduced, improving model interpretability and generalization ability, reducing the risk of overfitting, and enhancing the accuracy and robustness of named entity recognition.

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

Abstract

The application discloses a weight network reasoning method based on dynamic entity prototype, and aims at the problems of few data sources of word embedding expansion, only paying attention to high-dimensional features of entity words and ignoring semantic information of entity words themselves, inaccurate context information and insufficient model generalization performance in named entity recognition. The application carries out preprocessing operation on an entity recognition dataset; designs a prototype extraction algorithm, trains a model capable of generating dynamic prototypes; initializes all parameters of an entity recognition algorithm; deploys the extracted dynamic entity prototypes to the entity recognition algorithm, and iteratively learns from texts and entity prototypes by using a weight network and a multi-head attention mechanism to obtain a trained entity recognition model; and predicts entities of different datasets by using the trained entity recognition model to test the model performance. The technology can improve the problems of not effectively utilizing semantic information and poor generalization performance in entity recognition, thereby improving the accuracy of named entity recognition.
Need to check novelty before this filing date? Find Prior Art