Research On Law Article Prediction Method For Legal Texts

In recent years,in the field of intelligent justice,the analysis of legal judgment documents and the prediction of legal judgment based on the description of case facts have become a hot research issue in computational law.Prediction of legal judgment mainly includes the task of crime,law and sentence prediction.The purpose of law prediction is used by the fact description in the case and the law and regulation to predict the corresponding law of the case.The related work is mainly based on machine learning,neural network and other models,and predicts the corresponding law of the case by imputing the fact description in the judgment document.At present,the research has the following difficulties:(1)the task of rule prediction is a typical one to many problem,a single case may involve multiple rule,and the differences between different rule are usually small,the content and structure similarity of fact description is very high,the traditional method can not deal with this problem well.(2)criminal facts in the judicial documents are the true and objective description of the case,and the digital data included in them have an important influence on the distinction of legal articles.However,the current judicial judgment prediction model is not sensitive to the size of money,age and other figures,and lacks the analysis and processing of such data.Moreover,the current model applied to judicial judgment prediction cannot effectively obtain legal documents long-distance dependence of Ben.(3)At present,many research tasks in the field of intelligent justice have proposed the solution based on deep neural network,but the model generally has the defect of "black box",and the research process and experimental results usually lack the interpretab ility and reasoning mechanism.Take into account this,this paper proposes a prediction model based on model fusion and threshold filtering to solve the problems of small differences and one to many.Then,by analyzing the fact description and digital data(including money,age,etc.)in the correlation rule,a prediction model based on data discretization and deep pyramid convolution neural network(DPCNN)is proposed.Finally,in view of the lack of interpretability and reasoning mechanism in the research,an interpretable causal model for judicial decision reasoning is proposed.The relevant experimental results and case studies show that the method proposed in this paper improves the accuracy of the task,proves the importance of data discretization in the task of rule prediction,and supplements the lack of reasoning mechanism and poor interpretability of the existing judicial decision reasoning methods.The main work of this paper includes:Firstly,a prediction model based on model fusion and threshold filtering is proposed.By analyzing the fact description of legal documents and the specific judicial interpretation of legal articles,mining the characteristics of the fact description part of judicial documents,based on the open data in the "China Law Research Cup" judicial artificial intelligence challenge,a number of groups of experimental data sets of different sizes are constructed,and a number of groups of experiments are carried out on different data sets.The experimental results show that the proposed method can effectively improve the accuracy of the task compared with a single prediction model,and can better solve the recommendation problem of a single case factual description corresponding to multiple prediction models.Secondly,a prediction model based on data discretization and DPCNN is proposed.Combined with the contents of relevant laws,judicial interpretation documents and criminal law judgment documents,the digital data(including money,age,etc.)in fact description are systematically reprocessed.According to the particularity of legal forecast and the long-distance dependence of legal text,the data discretization method is applied to DPCNN for legal predict.Experiments have been done on multiple data sets.Compared with fresh latest baselines,the experimental results of this method are significantly improved.Thirdly,an interpretable causal model for judicial decision reasoning is proposed.Depending on the trial procedure and the theory of causal logic,this paper explores the discovery mechanism of causal relationship in judicial decision,and puts forward the construction mode of causal relationship diagram and the reasoning causal model of judicial decision(JCM).Based on the marriage documents and the relevant articles of marriage law of the people's Republic of China,this paper defines the judgment reasoning triplet of marriage cases < R,S,F >.This paper puts forward the verification strategy of the causal model of judicial decision reasoning,which proves that the model can better adapt to the court case decision process,has the objectivity,standardization and flow of judicial decision tasks,and can effectively supplement the lack of reasoning mechanism and poor interpretability of the existing judicial decision reasoning methods.