Research On Case Semantic Retrieval Based On Case Facts Knowledge Graph

In the context of the Internet era,as the national courts’s judgment documents are made public online,the public’s attention to justice is constantly increasing,and the judicial demand is increasingly strong.Equivalent judgments in similar cases are the people’s simplest understanding and cognition.Case retrieval is the only way to achieve the equivalent judgments in similar cases,the Supreme People’s Court has clearly established a retrieval mechanism for similar and related cases.Achieving accurate case retrieval will provide support for unifying adjudication criterion,maintaining judicial authority and assisting judges in adjudicating cases,and will have huge social impact and practical value.The existing case retrieval systems basically use the keyword-based retrieval method,which places greater demands on the ability of users to formulate keywords,and the retrieval results are often poor.This paper takes judicial documents as the research object,and use natural language processing as the core technology to conduct the frontier exploration and research on case semantic retrieval,hoping to improve the effect and experience of case retrieval.The main work of this paper is as follows:(1)The case facts knowledge graphs on large-scale judical cases are established.Based on expert knowledge,this paper defines the paragraph types of judgment documents and the data schema of knowledge graph;The formulated rules and strategies are used to structure the judgment documents;The models for entity and relation extraction are trained to achieve knowledge extraction for unstructured text based on the labeled data and the pre-trained language model;The automated construction process of the case facts knowledge graph is designed,and case facts knowledge graphs are established for hundreds of thousands of judgment documents.(2)A relation extraction model incorporating knowledge representation learning is proposed.In this paper,knowledge representation learning is introduced into the relation extraction task,and a model based on translational distance and a model based on semantic similarity are respectively incorporated.The results show that the effect of relation extraction can be significantly improved by using knowledge to guide and constrain,and the case knowledge embedding representation combined with context is obtained at the same time.(3)An entity relation extraction model based on joint learning is designed.Based on the existing relation extraction work,this paper designs a joint learning model for entity recognition and relation extraction.Compared with the pipeline-based method,the effect of the relation extraction task is effectively improved,and the underlying shared coding mechanism will improve the efficiency of knowledge extraction.(4)A prototype system of case retrieval based on case facts knowledge graphs is designed and implemented.In this paper,the overall architecture of the case retrieval system is designed;Based on the separated front-end and back-end architecture,the case semantic retrieval system is developed using case knowledge,and the case facts knowledge graph is visualized at the same time.The case application example shows that the system provides users with a good retrieval effect and experience.