Research On Construction Of Knowledge Graph Of Judicial Case Texts Based On Deep Learning

It is of great significance and application value to combine the rich,scattered and fragmented judicial knowledge together from the vast amounts of judicial case texts,which can provide users with accurate and effective services,thus to alleviate the contradiction between the increasing needs of people and limited high-quality judicial resources.The advantage of knowledge graph lies in its rich semantic processing capabilities and powerful organizing capabilities,which can analyze and depict the key points and internal connections of information and draws and displays the complex knowledge field with graphics.Therefore,this thesis utilizes judicial case texts in the judicial field as data sources to study the knowledge graph construction technology based on deep learning,which can not only help the users to quickly sort out the key information from the mass of judicial cases,but also provide relevant expanded information query function to assist users to make correct and reasonable decisions.In the stage of extracting information of the judicial case texts,under the background of the lack of publicly labeled corpus in the judicial field,the first task is to define the entity characteristics,location characteristics and speech characteristics and so on of the entities in the judicial case texts.Then on this basis,this thesis has developed the Bi_LSTM-CRF model for named entity recognition of judicial case texts.Firstly,the sentences in the judicial case texts are labeled by BIO notation and converted into character vectors as inputting.Secondly,the Bi_LSTM neural network is used to process the vector to obtain sentence features.Finally,CRF is used to label and extract entities to realize named entity recognition.At the same time,this thesis has also developed the Dependence Parsing Neural Network,named DPNN model to extract the relationships between entities of the judicial case texts.Firstly,the judicial case texts are pre-processed.Secondly,the dependency parsing module is used to deal with the sentence to get the dependency relationships between the components of the sentence.Finally,the sets of structured and effective entity relationship triples are obtained through the extract triples module.Experimental results show that the Bi_LSTM-CRF model and DPNN model developed in this thesis are both effective and feasible.In the stage of drawing the knowledge graph of judicial case texts,this thesis imports the entity relationship triples obtained in the previous stage into the Neo4 j graph database in batches,and realizes the visual display of the knowledge graph of judicial case texts through the Neo4 j graph database.On this basis,the extended query function is implemented to complete the task of constructing the knowledge graph of the judicial case texts.Experimental results show that the proposed construction method of the knowledge graph of judicial case texts can effectively obtain structured information from unstructured judicial case texts in the judicial field,and obtain entities and relationshipsbetween entities to completely construct knowledge graph of judicial case texts.It has laid an important foundation for the follow-up application of judicial information such as search engine,intelligent judicial question and answering,and intelligent judicial auxiliary decision-making and so on.