Research on Relation Extraction Based on BERT with Multifaceted Semantics

Relation extraction is one of the important tasks in natural language processing, aiming to determine the class of relations to which the entities in a sentence belong. In nowadays, researchers tend to use large-scale corpus to retrain language models for relation extraction, which requires too much relevant resources, besides, when analyzing the corpus, we found that in sentences where entities exist, different entities have different correlations with different parts of the sentence, and this may affect the performance of relation extraction. Based on this, this paper proposes a model based on BERT [1] with multifaceted semantics (BERT-LR) for relation extraction, which learns semantics from multiple aspects centered on entities, and is able to better understand the relationship between entities at different locations and the context and perform relation extraction. First, we make full use of the BERT [1] model that has been pre-trained to provide rich initialization parameters for our model. Second, in order to achieve entity-centered relationship extraction, we propose a BERT [1] and multifaceted semantic relationship extraction model based on BERT [1] with multifaceted semantics consisting of left semantics, right semantics, and global semantics, and use a suitable method to fuse the multifaceted semantics. Third, we found that fixing the embedding layer of the model during the fine-tuning process can achieve better results. Our approach achieves excellent results on the SemEval-2010 Task 8 [2] dataset.