Research on Tag Method for Joint Extraction of Domain-oriented Entity and Relation
Yuxin Shi, Ailian Zhou, Xiaohe Liang, Nengfu Xie, Saisai Wu, Xiaoyu Li · 2021
Corpus tagging is an important basic work of named entity recognition and relation extraction based on deep learning, and it is also a core task in natural language processing. Aiming at the problems of error propagation, information loss and entity redundancy in the pipeline method, this paper proposes a text annotation method "E+R+BIES" for joint extraction of domain entity relations based on the characteristics of specific domain corpus. The entity relationship extraction problem is transformed into a sequence labeling problem. The entity and its relationship information are included in a labeling process, the triples are directly modeled, and the BERT-BiLSTM+CRF end-to-end model is used for label prediction. Experiments show that the joint extraction effect of domain entity relations based on this annotation method and the BERT-BiLSTM+CRF model is better than the general frontier model, in which the accuracy rate is increased by 5 to 7 percentage points, the recall rate is increased by 9 to 11 percentage points, and the F1 value is increased by 8 ~9 percentage points, reaching 90.60%.