Method for Extracting Legal Elements Based on Judgment Text
With the continuous development of national court informatization construction,a large amount of judicial judgment data has been accumulated.How to extract accurate legal elements from legal docu-ment data is an important prerequisite for ensuring court informatization construction.Based on the legal document data of theft cases,this study constructed a BERT+BiLSTM+CRF fusion language model to solve the problem of extracting key elements from legal documents.BERT language model was used to solve the problem of polysemy in text feature representation,BiLSTM neural network was utilized to fully learn the characteristics of contextual information,CRF machine learning method was used to extract the global optimal annotation sequence,and a visual interface was established to provide case element extrac-tion services.The results showed that,overall,the BERT+BiLSTM+CRF fusion language model con-structed through data augmentation achieves a comprehensive evaluation index value of 90.6％for theft cases;From the extraction results of individual elements,the comprehensive evaluation indicator(F1_score)for the ten legal elements of the theft case were all above 81.8％;From the proportion of legal ele-ments distributed for optimal predictive performance,the model achieved optimal predictive performance with 50％of legal elements,which is significantly better than other models.This research indicates that the BERT+BiLSTM+CRF fusion language model can effectively solve the problem of extracting key ele-ments of judgment text,and provide a certain theoretical basis and effective technical support for the in-formatization construction of national courts.
万方数据
维普
