A Legal Question Answering System Based on BERT

With the development of artificial intelligence technology, intelligent question-answering systems in general fields have been widely accepted by people. However, the development of intelligent question-answering systems in limited areas is not very satisfactory. Moreover, due to the diversification of Chinese expressions, matching user input problems with prior problems is very important. This paper proposes a scheme to obtain the problem vector representation based on the BERT model. In addition, the Milvus vector search engine is used in this paper, which can not only provide store vector representation information but also calculate vector similarity. Finally, we return the answer through the database. When the threshold value of our proposed scheme is 0.2, the recall rate reaches 86%, and the mismatch rate reaches 84%. The results verify that the system has relatively good performance.