Research on Entity Relationship Joint Extraction Technology Based on Multitask Learning

Abstract

With the rapid development of information technology in recent years, entity relationship extraction as a key step of information extraction has been widely concerned. In the past practice relationship extraction technologies, entity identification and relationship extraction were mostly realized as two independent tasks, ignoring the relationship between the two tasks. In this paper, an entity relationship extraction model is proposed. Based on the pre-training model BERT, an entity identification and multi-head selection task model is constructed to jointly extract entity relationships. The model realizes entity identification through CRF, and classifies relationships by taking relationship extraction as a multi-head selection task. The joint extraction model proposed in this paper improves the tasks of entity recognition and relationship extraction. This paper integrates the entity relationship extraction model into the knowledge map management system, labels the entity relationship through the model, verifies the data through domain experts, and establishes the domain knowledge map. At the same time, the verified data is used as training data for incremental training of the model, so as to continuously optimize the model.