A Chinese named entity recognition model incorporating recurrent cell and information state recursion.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Chinese Named Entity Recognition (NER) faces challenges with syntactic complexity and limited long sequence feature extraction.
• Joint learning models improve NER accuracy but often overemphasize local features, hindering long-range dependency capture.

Purpose of the Study:

• To propose a novel Chinese NER model, IRCSR-NER, that overcomes limitations in long sequence feature extraction.
• To enhance the recognition of long entity boundaries and improve overall NER performance in Chinese.
• To incorporate syntactic dependency information to better handle Chinese language nuances.

Main Methods:

• Developed the Incorporating Recurrent Cell and Information State Recursion (IRCSR-NER) model.
• Integrated recurrent cells and information state recursion for improved sequence feature extraction.
• Utilized a syntactic dependency approach to add lexical relationship information at the word level.

Main Results:

• The IRCSR-NER model demonstrated superior performance in extracting long-text features and capturing long-distance dependent information.
• Experimental results on four public datasets showed that IRCSR-NER outperforms traditional and mainstream Chinese NER models.
• The model effectively balances local feature focus with long-range contextual information.

Conclusions:

• The proposed IRCSR-NER model significantly enhances Chinese Named Entity Recognition capabilities, particularly for long and complex sequences.
• Integrating recurrent cells and syntactic information offers a promising direction for advancing NER in syntactically rich languages like Chinese.
• IRCSR-NER provides a more efficient and accurate solution for Chinese NER tasks.