Question Answering Model Based on Graph Knowledge and Entity Recognition

With information increasing dramatically, how to make computers answer automatically and accurately questions, such as " What is the theme of Schindler 's List", has been becoming a hot research topic. However, the profoundness of language and culture often results in the problem with many forms of expression, which brings great challenges to our research. To overcome these difficulties, we use a triple form (subject, relational, object) to represent a fact and store it in the graph database for generating a large multi-relational graph. Among all kinds of questions raised by users, we only need to identify accurately the problem subject and map it to the graph database. For the returned series of candidate relationships, the relationship with the highest similarity to the problem pattern is calculated and the final answer can be obtained. The model includes two parts: entity identification and relationship prediction. In order to improve the accuracy of entity recognition, a BERT-based Sentence vector training and BiLSTM-CRF annotation entity recognition method are proposed.