| With the rapid development of big data and artificial intelligence,all industries are undergoing intelligent changes,and the field of judicial practice is no exception.However,the high specialization and large amount of case-processing work are real challenges to the development of judicial intelligence.The modern legal system has corresponding legal rules and procedures for each type of case.Therefore,it is crucial to determine the cause of actions of cases accurately.Traditional legal case classification requires a lot of time and workforce.The 14th Five-Year Plan explicitly states that we should "strengthen the construction of smart courts" and "promote the construction of universal and industry-specific artificial intelligence open platforms."Currently,natural language processing(NLP)is one of the most promising technologies in artificial intelligence,and the core task is to enable machines to understand and process natural language.Using NLP text classification technology,the algorithm of cause classification of legal case text based on artificial intelligence can be studied to automate the process of cause classification and greatly improve judicial efficiency.In addition,methods based on NLP and preprocessing automatically analyze case facts and legal texts,avoiding the influence of human factors on cases,thus improving judicial fairness and promoting the process of judicial intelligence.Therefore,it is of great significance to study the text classification techniques of causes of actions of cases in the legal field based on natural language processing technology.This thesis is dedicated to studying the key techniques of legal case text classification,combining natural language processing,dependency syntax analysis,and graph neural networks.In this thesis,new methods are proposed to improve the shortcomings of existing techniques.The study of this thesis includes the following three main aspects:(1)This thesis proposed an algorithmic framework for legal text classification by graph neural network combined with the enhancement of dependency syntactic information,aimed at the problems of sizeable legal text length,high professionalism,and technicality.The model can analyze the case text and extract the behavioral information of the parties.At the same time,it models the semantic relationships among document,behavior,and word information,and constructs an information-enhanced "document-action-word" three-level heterogeneous text graph.And uses two layers of Legel Text Classification,Dependent Syntactic Analysis,Graph Neural Network,Cause of Action Classification graph convolutional networks to achieve text classification.Finally,the effectiveness of the algorithm is verified on the self-constructed civil law case classification dataset.(2)Based on Work 1,this thesis proposes an improved deep graph neural network text classification model to solve two problems with existing models:(1)insufficient mining of text graph information;(2)inability to learn semantic relationships between different types of nodes and edges.The model uses an improved graph convolution strategy to mine richer semantic knowledge between different types of nodes and edges,as well as to build a deeper graph network.To handle the rich edge types in the text graph,we use an improved convolutional computation and introduce the PairNorm layer and JK-Net to integrate intra-layer and inter-layer features to alleviate the limitation of GNN layers.Finally,the performance of the constructed graph network is verified by comparing it with the model in Chapter 3 on a self-constructed civil law case classification dataset.(3)Based on Work 1 and Work 2,this thesis designs and implements a system for case cause classification.The system adopts a C/S architecture,sends the case information uploaded by the user to the server for inference calculation,and finally realizes the classification of the case cause category and the recommendation of related legal provisions. |
PDF Full Text Request