Research On Legal Articles Prediction And Similar Case Matching For Judicial Big Data

In recent years,the analysis and mining of legal adjudication documents in the field of intelligent justice has become a hot research issue in computational law.The legal articles prediction task and the similar case matching task are two important tasks in intelligent justice.The legal articles prediction predicts the law involved in the case by analyzing the factual description of the case.The similar case matching selects the most similar case from the candidate cases by comparing the similarity of documents.The legal articles prediction and the similar case matching can assist judges in handling cases,and also help ordinary people understand the cases.At present,the research on intelligent justice mainly focuses on the prediction of judicial judgments,and there are few studies specifically on the prediction of legal articles and similar cases matching.The research on legal articles prediction and similar case matching for judicial big data is to use deep neural network model to realize legal articles prediction and similar case matching,and improve the level of intelligent judicial services.The main work of the thesis is:(1)In the legal articles prediction task,the convolutional neural network legal articles prediction model and the attention bidirectional LSTM legal articles prediction model are first constructed.In the experiment,the prediction results of the convolutional neural network article in different input layers and the Embedding layer are compared.Then,after analyzing the error examples,the confusing problem of the legal articles prediction is defined.Aiming at this problem,a confusing method of predicting the legal articles based on hierarchical learning is proposed.The model consists of a two-layer learning framework,with a convolutional neural network as a feature extractor,and a confusing legal articles model is trained separately.(2)In the similar case matching task,a similar case matching model of attention convolutional neural network is constructed.The model includes two convolution pooling layers,and an attention mechanism is added to theconvolution layer.Aiming at the problem that there is less training data in similar case matching and it is difficult to distinguish between documents,a twin Bert similar case matching model is proposed.The main frame of the model uses a twin structure,with Bert as the document coding network,and the similarity of the documents is calculated by the cosine formula.(3)The model is tested on real case data based on the Chinese referee document network.In the legal articles prediction task,the confusing legal articles prediction model based on hierarchical learning achieved the best results on both the CAIL dataset and the custom confusable method data set,indicating the effectiveness of the hierarchical learning model for confusing legal articles prediction.In the similar case matching task,the accuracy of the twin Bert similar case matching model reached 88% in CAIL2019_Small and78% in CAIL2019_Large,indicating that the twin Bert similar case matching model can better solve the problem of less training data and difficult to distinguish between documents in similar case matching.