Research On The Construction Technology Of Criminal Law Knowledge Grap

With the continuous improvement of Chinese judicial construction,a large amount of(semi-structured and unstructured)judicial big data is emerging on the Internet.There are various judicial data such as law-related encyclopedic knowledge,criminal law,judicial interpretations and judgment documents.It is important to make full use of these judicial big data to promote judicial openness,fairness and efficiency.The knowledge graph can link different types of knowledge through nodes and edges to form a huge knowledge system for users to query.And the knowledge graph itself also has a certain reasoning functions,which can infer new information based on the original knowledge.Therefore,this thesis studies the construction technology of criminal law knowledge graph based on judicial data in the criminal field to assist users in making reasonable decisions.The main work of criminal law knowledge graph construction technology are as follows:(1)Named entity recognition of criminal judgment documents.In order to improve the recognition effect of complex entities in judgment documents,a named entity recognition method based on BERT and joint learning(JLB-BiLSTM-CRF)is proposed.In this method,firstly,the input character sequence is encoded by BERT to enhance the representation ability of word vectors.Then,the long text information is modeled by the BiLSTM.Besides,the named entity recognition tasks and Chinese word segmentation tasks are jointly trained to improve the boundary recognition rate of entities.Experimental results show that JLB-BiLSTM-CRF has the F1 value of94.65% on the test set.It confirms the effectiveness of JLB-BiLSTM-CRF in named entity recognition task for judgment documents.(2)Relation extraction of criminal judgment documents.In order to improve the efficiency of relation extraction in judgment documents,a document-level Chinese relation extraction method based on BERT(MCR-BERT)is proposed.In this method,the input document is encoded only once by the improved BERT and relationship classification is determined by the relevant contextual information of the target entity pair.Experimental results show that MCR-BERT can effectively shorten the training time of the model and achieve a better relation classification result.(3)Construction of criminal law knowledge graph.In order to integrate a variety of judicial data to expand the use of the legal knowledge graph,firstly,a rule template is constructed by analyzing the characteristics of the four judicial data to extract structured knowledge.Secondly,a method based on Word2 vec cosine similarity is used to align entities with various expressions to eliminate knowledge redundancy.And then Protege and Jena inference engine are used for ontology construction and reasoning of criminal law knowledge graph to generate new triples.Finally,the generated triples are stored in a Neo4 j graph database.The overall constructed criminal law knowledge graph contains 55390 entities and 176292 relations.