Similar Case Retrieval Method And System Based On Speech Question And Answer

Judgment documents are the most crucial carrier in legal practice,but from them,we often see different sentences for the same case,which severely impacts on the judicial credibility and fairness.Therefore,it is crucial to check the sentences of similar cases.Although a few systems of this kind have been developed,they do not meet well the needs of real legal practice.Moreover,most of their methods are based on keyword matching,which cannot find similar cases according to the criminal facts described in a natural language,which leads to the incomplete reference of similar cases,and further leads to the different judgments of similar cases.The possible reasons for these problems are as follows.1)It is not accurate enough to meet a judge's actual needs.2)The scope of similar cases retrieval is too narrow,the source is not clear,and the levels of different courts are not clear enough.And 3)local autonomy of courts leads to significant differences in similar cases judged in different regions.To the end,in this paper,we develop a similar criminal case retrieval system via multi-round speech question and answer(Q & A)based on text similarity.More specifically,our system uses the techniques of online speech synthesis of Chinese IFLYTEK and natural language process to recognise users' questions in speech and answer them in speech and text.Moreover,we propose a text similarity calculation method to retrieve the most similar cases based on their structure and semantics.Our experiments carry on 1,000 real judgments and show that the system has the high level of human-computer dialogue function and high accuracy of information retrieval,which can well satisfy the needs of similar case retrieval in legal practice.The main research work of this paper is as follows:1.The structuring process of unstructured judgment text data.In this paper,we use different algorithms to extract information from different entities in legal decision documents,such as wordnet-based algorithm,stanford-based entity extraction algorithm,rule-based method,thesaurus based method,regular-based method.2.We use the existing semantic slot technology to set the semantic slot library and synonym library for the legal field.And on the basis of the existing semantic slot technology,we improve the answer semantic part of the semantic slot by adding the answer semantic information of the rhetorical question,so that our system can process the context information and realise themulti-round dialogue needed.3.According to the existing law of similar cases demand,structured way can realise the court and the court's preferred is recommended.However,due to the complexity of the legal text,the structured data cannot fully reflect the nature of the crime process,so this paper proposes to combine the text similarity value with the structured text weight.The statistical analyses of the experimental results show that the performance of this model out performs the state-of-the-art algorithm.4.In terms of speech research,IFLYTEK has developed rapidly in recent years,and its speech recognition and synthesis technologies are relatively mature.In order to provide convenience for users,the system in this paper incorporates the speech recognition and synthesis technology developed by IFLYTEK,which enables users to interact with the system by means of speech input and speech output.