Research On Key Techniques Of Legal Consultation And Auxiliary Judgment Based On Tensor Decomposition

In recent years,the Ministry of Justice has fully implemented Xi Jinping Thought,focused on improving the level of judicial administration,carried out scientific and technological innovation and technological breakthroughs,and built the "digital rule of law,smart justice" scientific and technological innovation project.Legal consultation service and auxiliary adjudication are important items among them.At present,legal consultation services is not intelligent enough to describe legal consultation sentences accurately.It is difficult to understand legal consultation intention of users.Traditional auxiliary judgment technique has a coarse-grained description of cases,which cannot capture correlation information between cases and affect legal conviction.In addition,for traditional auxiliary judgment algorithms,it is difficult to extract effective information accurately from cases and obtain detailed pre-judgment results.This thesis makes an in-depth study of the above issues,trying to accurately depict legal cases and counseling sentences.It predicts users' intention of counseling,realizes the conviction of case and prediction of judgments.The main contributions of this thesis include:First,we study the unified and standardized representation of judicial cases.Judicial case modeling provides data support for case-related prediction algorithms.Traditional case modeling methods are mainly based on feature models and matrix decomposition.These methods have the natural flaw of feature models.In response to these problems,a judicial case modeling method based on normalized tensor decomposition is proposed.The concept of tensor model is introduced to express cases as three-dimensional tensors.Then a weight matrix is defined.The normalized tensor decomposition algorithm is used to convert original tensors into core tensors and complete the judicial case representation.Core tensors reduce the dimension of original tensors.They remove redundant information and reduce the computational complexity of subsequent prediction algorithms.Compared with traditional methods,our method significantly improves the accuracy of prediction algorithms on real legal consultation sentences and legal case datasets.Second,we study the method of understanding the intention of legal consultation.An accurate understanding of users' legal consultation intentions is a prerequisite for providing personalized legal services.Modeling methods and prediction algorithms in traditional methods on intention understanding of legal consultation are independent of each other.Prediction algorithms cannot accurately extract effective information from legal consultation sentences.In response to above problems,a method based on pattern tensor decomposition is proposed.It optimizes the unsupervised tensor decomposition.The pattern tensor is defined.On this basis,tensor decomposition algorithm is used to express legal consultation sentences as core tensors.An optimization method of prediction algorithms with respect to pattern tensor is designed.In turn,prediction algorithms intervene in tensor decomposition process by optimizing pattern tensor.Pattern tensor establishes the association between modeling process and prediction algorithms,so that core tensors represent the tensor element and structure information that is most conducive to improving the accuracy of the method.Compared with traditional methods,our method effectively improves the accuracy of consultation intention understanding on the real consultation sentence dataset.Third,we study the method of determining crimes in cases.The conviction of legal cases is vital to the triage of cases and the distribution of judges,which directly affects the speed of case circulation and the efficiency of trial.Traditional methods cannot take similar information between cases as an important basis for predicting crimes in cases.In response to this problem,a method based on similarity-driven neural networks is proposed.On the basis of pattern tensor decomposition,it improves the gate control structure of neural networks and defines input and output similarity gates.The two respectively obtain similar information between cases in the input and output layer of neural networks,and use it in the calculation of final results.The similarity gates capture potential associations between cases.They provide similarity information for prediction algorithms,and effectively improves the accuracy of judging crimes.Compared with traditional methods,our method has a higher accuracy on the real legal case document dataset.Finally,we study the method of case outcome prediction.Prejudgment of cases is helpful to promote the trial of cases,assist judges to judge cases,and reduce the occurrence of unjust,false and erroneous cases.Traditional methods are mainly based on classification algorithms,and the particle size of prediction results is rough.In addition,modeling process has poor correlation with the prediction algorithm,which cannot accurately obtain effective information from cases.In response to above problems,a method based on intervenable tensor decomposition for is proposed.It optimizes pattern tensor decomposition.It defines intervenable parameters,constructs controllable and mapping tensor decomposition algorithms,which use intermediate tensors and mapping matrix sets respectively to optimize tensor decomposition process.The optimized regression model is designed,so that core tensors represent the tensor information that is most conducive to improving the accuracy of prediction algorithms.Compared with traditional methods,our method has a significantly higher accuracy of judgment prediction on the real legal case dataset.