CFin-NER: A Study on Chinese Financial Entity Recognition Based on Comparative Learning

Jigui Zhao, Yurong Qian, Shuxiang Hou, Jiaying Chen, Kui Wang

Abstract: With the rapid development and digital transformation of the financial industry, financial text data has grown dramatically. Among these, Chinese bank news has become a key information source for analyzing financial markets, predicting risks, and formulating policies. However, the generic NER model performs poorly in the financial domain. Therefore, this paper proposes the CFin-NER model for the specificity and complexity of named entity recognition in the financial domain. It is a FinBERT-BiLSTM-CRF framework based on comparative learning. Firstly, we construct a high-quality Chinese bank news dataset, Fin_Bank, covering rich financial entity annotations. Secondly, FinBERT, a pre-trained model designed for the financial domain, is augmented by introducing a contrastive learning strategy that enables it to understand financial terminology and complex financial information more deeply and capture the dependencies of long-distance named entities. Finally, the output of FinBERT is passed to the BiLSTM layer to capture contextual information and decoded by the CRF layer to achieve accurate recognition of financial named entities. Experimental results show that the model proposed in this paper achieves significant performance improvement on the Fin_Bank dataset.

External IDs:doi:10.1007/978-981-96-6963-9_31