Research And Application Of Similar Case Matching Method Combined With Case Element Sequence

Similar case matching is to find cases with the same criminal circumstances as the specified case from the case database,thereby supporting intelligent auxiliary case handling such as automatic sentencing and legal recommendation.Traditional methods are mainly based on the bag-of-words model to represent cases as vectors,and obtain similarity by calculating the distance between the vectors.However,in the actual trial process,the case elements and their orders have a direct impact on the measurement of penalty.Due to the lack of consideration of case element knowledge in traditional methods,the semantic analysis is weak,and it is hard to support the case’s matching application.In response to the above problems,this paper proposes a similar case matching method that combined the sequence of case elements.This method uses information extraction technology to extract case elements automatically,constructs the sequence of case elements,then abstracts representation based on the Graph Convolutional Neural Network,and integrates text semantics for similarity calculation,thus supporting practical applications.The main work of this paper divided into the following three parts:(1)Construction of case element sequence method.The existing methods lack case elements and weak semantic analysis,this paper combines Chinese Predicate Head and syntactic element identification methods,uses the BERT-CRF model to extract the case elements,and constructs the sequence of case elements.Utilizing the temporal relationship between the elements,modeling semantic dependence and structural features,so that achieve a deep understanding of case semantics and effectively support the matching of similar cases.(2)Similar case matching method.To effectively obtain the semantic dependent features in the case element sequence,the Graph Convolutional Neural Network is used to model the case element sequence,the judgment result is used as the label to training model to learn the abstract representation of the case element sequence,and then fuses the text semantics.The case is expressed from the different dimensions of case elements and text semantics.Finally,the result of crime prediction is used to match similar cases.Experimental results show that this method can significantly improve the similar matching performance and find similar circumstances cases.(3)Similar case matching system.Based on the research of similar case matching methods,this paper designs and implements a system,which includes functions such as case element sequence construction,case retrieval,and similar case matching.The system realizes the preliminary practice of similar case matching,and effectively supports the application requirements of intelligent auxiliary case handling based on similar case matching.