Research And Implementation Of Knowledge Graph For Judicial Administration

With the development of knowledge graph technology research and its application in all walks of life,the application research in the field of justice and administration is still in its infancy compared with the application in the open market.The main problem to be solved in this paper is how to construct a judicial administrative knowledge graph with unstructured data under a large number of structured data of different standards,and how to correctly analyze users’ natural query statements and meet users’ search needs under different search habits.The main work of this paper is as follows:Firstly,this paper proposes a method to quickly build a knowledge map based on structured data.Due to the accumulation of a considerable number of judicial and administrative databases in the project,the content quality of these databases is very high,but these databases can not be directly used to build a knowledge map.We need to think of ways to build relationships between different tables.After investigation,this paper puts forward a structured data map based on voting Relationship discovery and extraction methods,so that these data can be used.Secondly,we propose a relationship extraction model of judicial documents based on BiLSTM-Attention and probability graph.It is still insufficient to extract the relationship only by relational database,or we need to extract the relationship from a large number of judicial documents.The task of relation extraction can be understood as a classification task,and the content of the knowledge map can be greatly supplemented by relation extraction of unstructured text.Thirdly,this paper proposes a semantic parsing method of query sentences.Combined with cypher’s syntax features,it selects the top-down syntax tree method to solve this problem through the seq-to-seq framework.At the same time,it uses slot filling method to make up for the poor parsing effect when the available training data is insufficient at the initial stage.Fourth,according to the search characteristics of the judicial field,advanced search instructions are specially designed to enable users to obtain relatively high quality search results with low learning cost.In addition,it also takes into account the keyword search method left by users when using traditional search engines,and makes some optimization for the judicial field and task characteristics.The combination of the three search styles which enables users to choose the right way according to their own habits and purposes.Fifth,this paper constructs a complete application system based on judicial administrative knowledge map,and puts forward an effective task effect measurement method according to the characteristics of the system.The application system runs in the Linux environment and adopts the strategy of separating business and interaction,which makes the subsequent maintenance cost lower and the deployment more convenient.