Research On Machine Reading Comprehension Technology For Legal Field

With the development of artificial intelligence technology,the application of artificial intelligence in the judicial field has gradually faded in people’s sight.In order to combine with the needs of smart justice putting forward by the Supreme People’s Court,the research goal of this thesis will focus on researching and developing an algorithm that can read and understand legal judgment documents and can answer questions related to legal judgment documents.To achieve this goal,this thesis makes the following contributions:1.In order to solve the problem that the legal text is too long and lacks long-term dependence and the encoding vector of the text and the problem is low in correlation,this thesis proposes a legal text reading comprehension method based on a multi-attention mechanism.The combined use of the mechanisms not only solves the problem of long legal texts,but also solves the problem of low text relevance.At the same time,the multitask joint training method is used to solve the extraction of various question types and question clues in legal question answering with one model.Experiments show that Ans＿F1,Sup＿F1 and Joint＿F1 are improved respectively by using multiple attention mechanisms than the other attention method.2.In order to solve the problem that existing methods do not work well for problems that require multi-step reasoning,this paper proposes a multi-hop reading comprehension method for legal texts based on asynchronous hierarchical graph neural networks.The algorithm proposes a hierarchical graph network,which includes four levels: sentencesentence,sentence-entity,entity-entity,and entity-sentence.And follow the logic of human reasoning,asynchronously update the hierarchical graph network to obtain the reasoning path.Experiments show that compared with the previous multi-hop reading comprehension method and the method in Chapter 3,the use of this algorithm has significantly improved the experimental effect.3.In order to prove the practicability and effectiveness of the algorithm proposed,a legal consultation system is designed in this thesis.The algorithm part of the system is integrated into the legal text reading comprehension algorithm based on asynchronous hierarchical graph neural network proposed in this subject,and the design of the system web page is completed.Implemented a system that can automatically answer user questions according to legal texts,and supplemented the question-and-answer data in the background database,so that the system also has the function of free consultation.