Statutes Recommendation Algorithm Based On Neural Networks

Along with the vigorous development of artificial intelligence and the extensive application of judicial intelligence,by reading a large number of cases through the machine,the system automatically extract the relevant legal statutes involved in the case,and realize statutes recommendation Intelligently.Furthermore,the statutes recommendation era help non-legal professionals who are not familiar with complex statutes quickly understand the penalties that they may face according to the facts of relevant cases.Base on the wide applications of automatic statutes recommendation,this thesis,by comprehensively analyzing the Chinese text descriptions of legal cases and the contents of the statutes,while also takes the characteristics of the task statutes recommendation into consideration,researches on the statutes recommendation method for criminal cases based on neural network.The work of this thesis mainly includes the following three aspects:(1)Statutes Recommendation Using Classification and Co-occurrence Between Statutes.This is the baseline model of this thesis.The main implementation method is to regard the statutes recommendation task as a Multi-label classification task.Through the analysis of a large number of criminal cases,the cause of action is concluded that criminal cases can be divided into six different categories.With a co-occurrence relationship between different statutes.The model uses TF-IDF to create document vectors.It's composed of a cause of action classifier and a stripe classifier to generate a set of candidate statutes.Finally,through the association rules between the statutes,the most likely reference to the law can be observed.(2)Attention-based Recurrent Convolutional Neural Network for legal Statutes Recommendation.To encounter the deficiencies of baseline system as ignorance of complete information keyword vector and the limited application of syntactic information,we propose a hierarchical sentence-document-level neural network model to obtain contextual and syntactic information.By jointly training multiple binary classifications,the improvement in computational efficiency as well as the implementation of rule task can be achieved.(3)Legal Statotes Recommendation based on Memory Network.With reference to the intrinsic dependencies between statute contents,a new legal recommendation model that integrates the memory network and the two-layer attention mechanism is proposed.The model,by means of external memory cell,captures the relationship between the statutes and the case description as well as the intrinsic dependencies between statutes.Finally,we can get more effective case-based and statute-based joint representations,as to send them to the classifier to implement statutes recommendation task.In the experimental verification stage,the effectiveness of the proposed legal stat utes recommendation algorithm is verified with a series of comparative experiments,a nd the possible future work directions are also expounded.