Research On Law Text Relation Extraction Algorithm Based On Knowledge Graph

The application research of relation extraction technology in the judicial field can assist judicial clerks to mark papers efficiently and clarify case information quickly.At the same time,knowledge graph can use knowledge to enhance the representation of text and provide interpretability for relation extraction technology.However,the combination of knowledge graph and relation extraction technology will inevitably meet their own problems in the judicial field.The existing knowledge graph in the judicial field is incomplete and the relationship between entities is missing.At the same time,knowledge graph completion algorithms based on text representation have achieved the most advanced performance,but they do not take into account entity neighbor information.In addition,in the knowledge graph,the attributes of each entity of triplet in addition to the ownership relationship also exist entity type.The existing relation extraction algorithm does not model the type,which leads to the problem of incomplete information utilization.More importantly,how to use knowledge graph to enhance the knowledge of legal texts has become a problem to be solved.In order to solve the above problems,this paper improves the knowledge graph completion algorithm and relation extraction algorithm respectively and applies them to the judicial field.The specific research content of this paper is as follows:(1).In view of the incompleteness of knowledge map in legal domain and the fact that the existing knowledge map completion algorithms based on text representation do not consider neighbor information,an improved knowledge graph completion algorithm based on translation model,graph neural network model and text representation based on contrast learning is proposed.Through a two-stage pre-training,the complex relationship information between entities and multi-hop neighbor information between entities in the knowledge graph are studied.The experimental results show that compared with other models,the proposed model improves at least 2.1% on Hits@10 score and 2.7% on MRR score,indicating that the proposed model has good performance.(2).An improved global pointer connector algorithm is proposed to solve the problem that the existing entity relation joint extraction algorithm does not fully utilize the triplet information.In the decoding side of the relational extraction algorithm,the reference detector in the original global pointer connector is replaced by two new global Pointers to identify the type of the head and tail entity,so as to make full use of the information of the knowledge graph.The experimental results show that compared with the unimproved global pointer connector,the F1 score is improved by at least 1.7%,which indicates the effectiveness of the improved algorithm.(3).Aiming at the problem of how to use knowledge graph to enhance the knowledge of legal texts,an algorithm based on knowledge graph to enhance the knowledge of legal texts is proposed.At the encoder of the relational extraction algorithm,the enhanced features are used in the improved global pointer connector,and the multi-head cross-attention module is used for fusion inside the model.The experimental results show that compared with other models,the F1 value of the proposed model increases by at least 0.16%,indicating that the performance of the improved algorithm is good.The research of this paper enriches the application of knowledge graph completion algorithm and relation extraction algorithm in the judicial field,and provides reference value for the direction of text relation extraction of knowledge graph in the judicial field,which has theoretical and practical significance.