Joint modeling method of question intention recognition and slot filling for domain-oriented question answering system

Question answering systems are developing continuously. To make the machine comprehend the user input discourse better, question processing has become an active research topic of natural language processing. Intention recognition is a semantic classification task and a crucial step in question answering system construction. The effect of intention recognition influences the response of the whole question answering system. Therefore, an efficient intention recognition model plays an important role in the construction of a question answering system. The intention recognition task is correlated with the slot filling task. We propose an improved joint modeling method for intention recognition and slot filling, in which the two tasks are executed simultaneously. The experiment is carried out on two public datasetsATIS and Snips, and the experimental results demonstrate that the proposed model has an optimization effect. Based on the computer literature Knowledge Graph (SCIKG), the domain-specific dataset named CSLQ is constructed. In addition, experiments on the previous joint task model verify the effectiveness of the constructed dataset CSLQ. The experimental results on the three kinds of datasets demonstrate the effectiveness of the proposed model compared with the other methods.