Research On Key Technologies Of Similar Case Recommendation Based On Legal Elements

With the proliferation of digitally stored texts of legal cases,legal workers and the general public review these materials more easily,but the retrieval of useful information from these texts also takes a lot of time,so accurately finding information is very meaningful,and similar legal case recommendation is an important part of it.This paper proposed an unsupervised legal text vector generation model based on graph embedding,which combines the legal element information and the semantic information contained in the legal case texts.This model first extract the legal elements contained in each case,and then generate a subgraph for each legal element dimension,every node represents each case,and every edge represents that two cases contain the same legal element.Further,the graph embedding model is used to generate a vector representation of each legal element dimension for each case,and finally the vector representations of different dimensions are concreted to form a separate vector to represent the text,and the vector is used to calculate the similarity of two legal cases.Considering that the importance of different legal elements is different,but the laws and regulations do not clearly stipulate the importance of different elements.Therefore,when concreting vector representations of multiple dimensions of legal elements,this paper uses the judgment results as labels training model to make the model are able to learn the weight information among different elements,and then this model uses these weight information to get the final vector representation of a legal text.Because the defination of similar cases has no strict regulations in the legal field,that is,there is no sufficient and necessary condition to define the similarity of the two cases,and the lack of authoritative similar case data sets in the field of criminal cases,the evaluation of models is difficult.However,similar cases are asked to have similar judgment results.Therefore,this paper used the judgment results to design a case similarity evaluation index for evaluating the model proposed in this paper.Finally,this paper tested on a data set and found that the similar cases recommended by the model proposed were superior to the traditional text similarity calculation method.