BERT-BIGRU-CRF: A Novel Entity Relationship Extraction Model

Abstract

Entity name recognition and entity relationship extraction are the most critical foundation for building knowledge graph, and it is also the basic task of NPL. The main purpose of entity relationship extraction is to extract the semantic relationship between the pairs of marked entities in the sentence, that is, to determine the relationship categories between entity pairs in unstructured text based on entity identification, and to form structured data for storage and retrieval. This paper proposes a BERT-BIGRU-CRF entity relationship extraction method, which effectively changes the relationship between the pre-training generated word vector and the downstream specific NLP task, and gradually moves the downstream specific NLP task to the pre-training generated word vector. Our method achieves better performance of relationship extraction and entity name recognition, which helps to construct the knowledge graph more accurately.