| With the promotion of legal construction,number of cases is rapidly increasing,problems such as " more cases,less people" and " unequal judgments in same cases" become more and more serious.How to effectively use the accumulated cases knowledge becomes an important subject of intelligent justice.Case text is the main carrier of case knowledge.By analyzing and understanding the case text,can help judicial officers obtain the key information of the case quickly,help archiving the case text and help judicial officers make reasonable judgments on new cases.Most of the current research on case text focuses on the prediction of criminal case text in terms of charge and sentence prediction,or entity extraction based on case text,but less research on the overall understanding of case text.For the task of Chinese case text understanding,this thesis conducts research on several key tasks such as generic representation framework of case text,case keyword discovery,case fault sentence analysis,and case class case discovery,establishes corresponding intelligent analysis model,so as to provide an effective technical support for intelligent assisted case judgment.The main work and innovation points are as follows:（1）By analysis of the cases text and text writing specification,we define a general case-text representation framework,and classify the elements in the framework into extractive case elements and analytical case elements according to the way of obtaining the case elements,which provides a knowledge guide for each key technique of subsequent case understanding.（2）Keywords reflect the case’s key information,which can help get the case’s key points and help find the relevant case text.Existing keyword extraction methods are not effective for extracting case text with multiple paragraphs or topics.Therefore,a keyword recognition model for case text is constructed by combining the characteristics of case text.The model integrates grammatical information from the case text,and introduces thematic elements to filter keywords,offsetting gaps in current methods.（3）The fault information is the main basis for case adjudication.Based on knowledge engineering,construct a "fault feature" dictionary and a fault sentence annotated corpus;based on neural network technology,design a short text classification model based on Bert and fully connected neural network to complete fault sentence recognition.（4）Classes of case texts are derived from the relevant definitions of laws at present,which can assist in law selection and text archiving.Existing classification methods cannot obtain effective information well when dealing with case texts,so the idea of reconstructing case texts based on extractive elements and then realizing classification is proposed,design the corresponding input construction method and long text classification model.（5）Referring to the judgments of similar cases can improve the accuracy of judgments,so design a Graph model to calculate the text similarity of cases based on the extracted case elements,and realize the class case pushing based on the text similarity.Finally,the data is visualized for the extracted elements,which helps the relevant practitioners to obtain the key elements of the case text more intuitively and analyze them. |
