Hybrid answer selection model for non-factoid question answering
Rongqiang Ma, Jian Zhang, Miao Li, Lei Chen, Jin Gang Gao · 2017
Capturing the semantic associations between questions and answers is a challenging task for answer selection. In this paper, a hybrid answer selection model is proposed by combining Convolutional Neural Network (CNN) and abstract extraction methods. In the model, answer summarization is extracted from the text with multiple features, and sent to the CNN together with the question to obtain a concise and efficient semantic representation. Unlike previous deep models, irrelevant information is removed and better representations are generated for question and answer, which is necessary for non-factoid question answering. The results on two datasets InsuranceQA and Agriculture QA show that our model outperforms other single deep models.