Research On Legal Document Representation Enhanced By A Knowledge Graph

With the development of informatization technology in the judicial field,the legal text,the legal text as the basic data in the judicial field has become increasingly important in the process of judicial intelligence.How to use these legal texts efficiently has very important research significance for the prediction of legal judgments such as crime prediction and legal recommendation.Most of the existing legal text representation methods are learned using neural network models.However,these methods have not considered some unique attributes of the judicial field text data,nor have they considered the massive knowledge contained in the knowledge map to help the legal text representation task.From the perspective of the legal text,the legal text itself contains many special words,and lacks an explanation for these words.From the perspective of knowledge graphs,the existing legal knowledge graphs mostly use cases,involved persons,judges and courts as entities to construct their existing relationships,and the knowledge contained in them is difficult to integrate with the legal text representation.Out of these characteristics of the legal text and the shortage of the existing legal knowledge map,this paper builds a legal knowledge map for criminal behavior and studies how to introduce the unique attributes of knowledge and legal text in the knowledge map into the neural network model for Generate legal text vector representations with more information.Specifically,the main research work in this article can be summarized as the following aspects:(1)Aiming at the requirements of the case representation in the text,a knowledge graph construction method for criminal behavior is proposed.Through the analysis of the case in the legal text,the LTP tool of Harbin Institute of Technology is first used to perform word segmentation,part-of-speech tagging,and named entity recognition.Then the semantic relationship between the entities is extracted and stored in the processed text through the dependency syntax analysis algorithm It is in the form of triples,and then the data information in the form of triples is input into the Neo4 j graph database,and Neo4 j is used to realize the construction of a knowledge graph for criminal behavior.(2)This paper presents a method of legal text representation that incorporates the characteristics of external criminal behavior.Through the analysis of the case in the legal text,the crime feature keywords in the case are analyzed,and related entities are queried in the knowledge graph based on these keywords,and these entities are converted into vector representations using the method of knowledge graph embedding.For the crime feature words,the word embedding method is used to transform into vector representation.In terms of model structure,the vector representations of the crime feature words and the corresponding knowledge entities are used as the multi-channel input of the convolutional neural network,and they are learned from the semantic and knowledge levels in the convolution process.So that the convolutional neural network can capture more information,and then can get a more complete vector representation.(3)This paper presents a method of legal text representation that integrates criminal behavior sequences.In generating the representation part of the criminal behavior sequence,the model extracts and stitches the criminal behavior described by the case by introducing multiple semantic paths in the knowledge graph to obtain the corresponding criminal behavior sequence,and uses a two-way long and short memory network to capture the sequence The semantic relationship of the text;in the generation of the text representation part,the text vector representation of the text,using the convolutional neural network to extract the local feature information of the text;finally,the results of the two models are combined to obtain the vectorized representation of the current text...