Research On Assistant Judgement Methods Based On Judgment Documents Of Natural Language Processing

With the continuous improvement of judicial transparency,the public has paid more and more attention to the judicial adjudication,which requires judicial officers to make more fair judgments.This paper studies the intelligent assistant methods of judicial adjudication that supported with a large number of judgment documents as experimental datasets,which uses the natural language processing technology to improve the work efficiency of judicial officers and help judicial personnel to conduct more fair cases.The main accomplishments are as follows:1.Based on the demand of intelligent assistant methods of judgement and the related technical research of artificial intelligence in the legal field,and using the judgment documents as the data resources,this paper proposes an assistant judgement scheme based on the combination of crime prediction and simi lar judgment document recommendation.It also introduces the theoretical knowledge involved in the assistant judgement method and the corresponding basic work,including: preprocessing of the documents data,building the case database,classifier training,manual labeling,etc.for the subsequent implementation of the assistant judgement methods.2.In the research of assistant method of crime prediction,it has low frequency word defect problem for the traditional chi-square statistical feature,and the feature subset selection process does not consider distribution uniformity of the feature item within the category,it tends to select the feature that is negatively correlated with the category,so that the frequency factor,the internal distribution uniformity factor and the negative correlation correction factor are introduced to improve these deficiencies,and the improved chi-square statistical feature selection Imp＿CHI proposed in this paper is applied to the crime prediction.3.In the study of assistant method of similar case document recommendation,the conventional TF-IDF model does not take the semantic information into account and the common Doc2 vec model usually ignores the influence of individual words on the whole document in the case text represent ation.A text representation method,Sif-Word2 vec,which combines word frequency and semantic relevance,is introduced into the recommendation of similar case document.4.Finally,the assistant judgement methods are proposed in this paper have implemented,and we designed the experiments for Imp＿CHI feature selection method proposed by the crime prediction assistant method and the Sif-Word2 vec model in the recommendation of similar case documents.the experimental results show that the best crime prediction results base on Imp＿CHI accuracy rate P is95.25%,recall rate R is 95.20%,F1 value is 95.22%,all are better than CHI and IG methods.In the similarity calculation and recommendation of the case document.In the similarity calculation and recommendation o f the case documents,the evaluation values of using Sif-Word2 vec model is 87% at Precision@5,and DCG@5 is 7.4432,both are higher than the evaluation values of using the TF-IDF models and the Doc2 vec models.According to the experimental conclusions,the improved method Imp＿CHI proposed in this paper has better feature selecting effects,which improves the accuracy of crime prediction.In the recommendation of similar case documents,the weighted model Sif-Word2 vec introduced in this paper takes into account the individual word frequency and semantic information,so that it works better in the text representation and improves the recommendation performance.Therefore,the assistant judgement methods proposed in this paper are feasible in practice,which are based on the judgment proposal given by the crime prediction and the trial references provided by the result of recommendation of similar case documents.