Chinese named entity recognition based on feature fusion

Abstract

The study addresses the problem of Chinese Named Entity Recognition (NER), which involves identifying and classifying entities within text based on their meanings. Despite advancements in deep learning, challenges remain due to the complexity of the Chinese language, such as flexible word formation, entity ambiguity, and boundary detection difficulties. This research proposes an innovative feature fusion method that enhances semantic understanding by incorporating external information. Using neural networks, particularly graph convolutional and transformer-based models, the paper presents two new approaches to improve Chinese NER: feature fusion and syntactic enhancement. These models, tested on datasets like MSRA, Resume, and Weibo, demonstrated significant improvements in accuracy—6% and 1.5% for Weibo and MSRA, respectively. The proposed methods address key issues such as the ambiguous nature of Chinese entities and the need for more accurate boundary recognition. The result suggest that integrating syntactic and lexical information improves NER performance, especially in complex, real-world scenarios. The study contributes to practical applications in machine translation, knowledge graphs, and text processing systems, offering a robust approach to enhance Chinese NER tasks.