Research On Judgment Prediction Algorithms Of Criminal Cases Based On Domain Knowledge Graph

Artificial Intelligence(AI)has become a national strategy profiting from big data.The publication of massive legal data has provided significant support for the application of artificial intelligence in the legal field.Judgment prediction of criminal cases aims to predict reasonable relevant legal articles and accusations based on the facts of the cases,which has become a hot topic in the field of "AI+law".The knowledge graph provides a formal representation of the real world and can be used as a knowledge base to provide a more machine-understandable form of data and to maximize the intrinsic value of legal documents.Several domestic scholars have constructed and published high-quality Chinese knowledge graphs,but the application of knowledge graphs in the "AI+Law" field is still in its infancy,and further research is limited as researchers can not find open-source legal domain knowledge graph on the Internet.In this paper,we construct a specific knowledge graph in criminal domain,and explore knowledge representation methods that can further improve the traditional machine learning text classification algorithms and deep learning neural network models to support efficient criminal judgment prediction task.1.Using the criminal documents crawled by China Judgement Online as a corpus,extract criminal entities and relations between these entities from two open general Chinese knowledge graph,CN-Probase and OwnThink,to construct a special knowledge graph of criminal domain.The knowledge graph has more than 870 thousand entities and 1500 thousand relations,stored in Neo4j,and is published on github(https://github.com/dquaner/LegalKG).2.Aiming at the application of knowledge graph in text classification algorithm,this paper proposes a text similarity calculation model based on the extracted knowledge graph,which integrates the concept information and semantic information between entities.The text similarity calculation model is applied in KNN and SVM for legal text classification,so as to improve the performance of these two machine learning algorithms in criminal judgment prediction task.3.Aiming at the application of knowledge graph in deep learning model,this paper proposes an improved knowledge representation learning algorithm based on TransE.Then the knowledge graph embedding(KGE)with multi-source information,as input layer,is applied to CNN and RNN models to solve the problem of criminal judgment prediction.Comparative experiment has shown that the domain knowledge graph we constructed is reasonable and effective.With Micro-F1-measure and Macro-F1-measure as evaluation metrics,the improved text classification algorithms and neural network models based on this knowledge graph both perform excellent on the criminal judgment prediction task.