Research on Chinese Text Entity Relation Extraction Based on CasRel
Youyuan Zhang, Xinchun Ma, Li Zhang · 2024
The task of Chinese text entity relation extraction is one of the important tasks in the field of natural language processing. In order to solve the problems such as transitive errors and relationship overlap in the entity relation extraction task, this paper proposes an improved CasRel cascade extraction model. On the one hand, it uses the Chinese-Roberta-wwm-ext pre-training model as the text encoder. This model adopts whole-word masking for pre-training and utilizes a large-scale dataset to enhance its understanding of the Chinese context. It also inherits the advantages of the RoBERTa framework in language modeling and feature representation. On the other hand, it incorporates attention mechanism into the second-stage process to enhance the vector representation of entity encoding. And it was validated on the public dataset DuIE, with an F1 value of 76.53%. It was also applied on the self-made hazardous chemical accident dataset.