Research And Application Of Intelligent Question Answering Method For Theft Cases

With the rapid development of big data and artificial intelligence technologies,the country has comprehensively promoted the construction of knowledge-centric judicial informatization to ensure that intelligent and autonomous services are provided to judicial personnel and the public.Among the existing intelligent services,legal intelligence question answering systems cover functions such as legal consultation,knowledge graph question answering,machine reading and comprehension,etc.They are served for the broad masses of grassroots and judicial personnel and have universal significance.However,due to the large research cost and strong professionalism of legal intelligent question answering technology,the existing intelligent question answering systems often only have a single question answering function,and it is difficult to provide comprehensive and high-quality question answering services.Therefore,the development of a complete legal intelligent question answering system has important research and application value.The legal intelligent question answering system has the problems of strong professionalism and insufficient related research.The court’s existing consulting channels are insufficiently manpowered,and it is difficult to cope with a large number of simple and repetitive consulting tasks.If you use the Internet for legal consultation,ordinary people usually cannot get professional responses.Introducing legal expertise to construct a Q&A database can further enhance the professionalism of legal consultation responses.At the same time,the existing search systems of courts mostly use keyword models,which are unable to perform semantic analysis on user queries,and are difficult to return accurate and fine-grained answers,which is not conducive to simultaneous judgments in the same case and reduces trial efficiency.Knowledge graph question answering and machine reading comprehension use deep learning technology to realize the semantic understanding of question,and can provide question answering services for structured and unstructured data,thereby serving judicial practice.But in the judicial field,there is very little research on intelligent question answering.Therefore,this paper combines the deep learning method to carry out the research and application of three intelligent question answering technologies: FAQ,knowledge graph question answering,and machine reading comprehension for high-incidence theft cases in criminal cases.The main work of the paper is as follows:(1)Implement the FAQ question answering model for theft cases: According to the professional knowledge involved in theft cases,a question answering database of frequently asked questions about theft cases was constructed,and question retrieval was realized by using the BM25 algorithm and the BERT-whitening model.In order to further improve the matching accuracy,a question interactive matching model based on BERT-FT is proposed.(2)Realize knowledge graph question answering for theft cases: According to structured information such as relevant laws and judicial elements of theft cases,a case knowledge graph based on Neo4 j was built.An entity recognition model based on BERT-CRF is constructed,and the entity is linked to the knowledge graph through the domain dictionary.Use the BERT-based answer path feature matching model to sort and filter the candidate subgraphs.(3)Realize machine reading comprehension question answering for theft cases: Combining the characteristics of judgment documents in theft cases,a judgment document search engine based on Elastic Search was built,and a machine reading comprehension model based on Ro BERTa-WWM was designed.In order to further improve the performance and application effects of the model,a data augmentation method for unanswerable questions based on Uni LM,a machine reading comprehension model optimization method based on transfer learning,and a model robustness optimization method based on adversarial training are proposed.(4)Design and implement an intelligent question answering system for theft cases: Based on the court’s application requirements for intelligent question answering systems,the above three intelligent question answering models are integrated and an online intelligent question answering system is realized.The system includes functional modules such as legal consultation on theft cases,knowledge graph question answering,machine reading comprehension,and automatic question answering robot.