A Multi-Task BERT-BiLSTM-AM-CRF Strategy for Chinese Named Entity Recognition

Named entity recognition aims to identify and mark entities with specific meanings in text. It is a key technology to further extract entity relationships and mine other potential information in natural language processing. At present, the methods based on machine learning and deep learning have bee...

Full description

Saved in:
Bibliographic Details
Published in:Neural processing letters 2023-04, Vol.55 (2), p.1209-1229
Main Authors: Tang, Xiaoyong, Huang, Yong, Xia, Meng, Long, Chengfeng
Format: Article
Language:English
Subjects:
Annotations
Artificial Intelligence
Complex Systems
Computational Intelligence
Computer Science
Context
Data mining
Datasets
Deep learning
Feature extraction
Labeling
Machine learning
Methods
Natural language processing
Neural networks
Performance evaluation
Semantics
Words (language)
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Named entity recognition aims to identify and mark entities with specific meanings in text. It is a key technology to further extract entity relationships and mine other potential information in natural language processing. At present, the methods based on machine learning and deep learning have been widely used in the research of named entity recognition, but most learning models use feature extraction based on word and character level. The word preprocessing of this kind of model often ignores the context semantic information of the target word and can not realize polysemy. In addition, the loss of semantic information and limited training data greatly limit the improvement of model performance and generalization ability. In order to solve the above problems and improve the efficiency of named entity recognition technology in Chinese text, this paper constructs a multi-task BERT-BiLSTM-AM-CRF intelligent processing model, uses Bert to extract the dynamic word vector combined with context information, and inputs the results into CRF layer for decoding after further training through BiLSTM module. After attention mechanism network, the model can learn together on two Chinese datasets, Finally, CRF classifies and extracts the observation annotation sequence to get the final result. Compared with many previous single task models, the F1 score of this multi-task model in MASR and people’s daily datasets has been significantly improved (0.55% and 3.41%), which demonstrates the effectiveness of multi-task learning for Chinese named entity recognition.
ISSN:1370-4621
1573-773X
DOI:10.1007/s11063-022-10933-3