Chinese named entity recognition method for the finance domain based on enhanced features and pretrained language models

Abstract

For some named entities in the Chinese finance domain that are long, with difficult to delineate boundaries and diverse forms of expression, we propose a method based on pretrained language models for named entity recognition with enhanced features. First, the method considers entity boundary delineation and entity classification as two separate tasks and learns enhanced Chinese character features by introducing a gating-based multi-channel attention mechanism to delineate financial entity boundaries on the basis of a pretrained language model. Then, the boundary demarcation results are input into the pretrained language model in the form of mask units for data enhancement. Subsequently, document-level entity-based enhancement features are introduced to construct a finance entity classification model. We experimentally identified the best-performing Chinese pretrained language model from several state-of-the-art models and then embedded it into our method to compare against other benchmark models. The experimental results showed that our model is superior to other benchmark models on the named entity recognition task in the finance domain.