Research On Judicial Documents Classification Method Based On Bayesian Network

The recommendation of similar adjudicative documents can help the dispute coordinators to analyze and clarify the focus of disputes,and to improve the quality and efficiency of dispute resolution.Judicial document feature extraction and classification are the basis of high quality recommendations.The key legal elements are essential as document features,while they are hard to be retrieved since most judicial documents contain much redundant information.Furthermore,the element correlations are vital to document classification as well.First,this thesis proposes an effective key legal element extraction method.It evaluates the expressive power of legal documents based on the average information entropy weight index,then the legal elements with high occurrence frequencies but poor expressiveness are effectively filtered out.Second,a civil adjudicative information network is proposed to model the key legal element and the correlations as a whole.The network construction algorithm is designed and implemented in detail.Since the traditional Word2 vec algorithm is easily trapped in local optimum for the context window size limitation,this thesis proposes a legal element correlation extraction algorithm based on Network Embedding.It effectively discovers the latent logical correlations among the legal elements,and an optimization algorithm of generating legal elements feature sequences is given as well.Last,this thesis proposes two civil adjudicative information network based Bayesian network structure learning methods are proposed,and constructs a reasoning model to classify the judicial documents.The experimental results show that the proposed methods and algorithms are more effective in judicative document classification than the existed schemes.