Classification Prediction And Case Recommendation Of Judicial Cases Based On Natural Language Processing

Artificial intelligence technology has developed by leaps and bounds in recent years,and its theoretical and practical applications have even covered the judicial field.As an auxiliary tool in judicial field,legal intelligence may help to improve judicial efficiency,promote legal equality and justice.Based on natural language processing technology,this thesis constructs a charge classification model and a case recommendation model for judicial cases.The charge classification model is to predict the charges given the description of a case,which may assist judicial practitioners to predict the charges of conventional cases and reduce their workload.We use Macro-F1 score as the model evaluation index,and SVM classification model as the baseline model,in order to compare the performance of the three multilabel classification models of the Text CNN classification model,BERT classification model,and Mean-Max BERT classification model.The result shows that Mean-Max BERT classification model performs the best among the three models.The judicial case recommendation model is designed to provide similar cases to judicial practitioners.Based on the cosine similarity between the input case description and the case description in the case database,a case with a similarity to the input case description of more than 50% was selected as the recommended case.Compare the TF-IDF recommendation model,Mean＿Word2vec recommendation model,Weight＿Word2vec recommendation model and Doc2 vec recommendation model through Micro-F1 score,we find that the Weight＿Word2vec recommendation model has the best recommendation effect among the four models.The case recommendation model is helpful for judicial practitioners to make correct convictions and sentencing judgments for non-conventional cases,and to achieve "same case,same sentence".