Bi-directional Capsule Network Model for Chinese Biomedical Community Question Answering

摘要

With the rapid development of the Internet, community question answering (CQA) platforms have attracted increasing attention over recent years, particularly in the biomedical field. On biomedical CQA platforms, patients share information about diseases, drugs and symptoms by communicating with each other. Therefore, the biomedical CQA platforms become particularly valuable resources for information and knowledge acquisition of patients. To accurately acquire relevant information, question answering techniques have been introduced in biomedical CQA. However, existing approaches cannot achieve the ideal performance due to the domain-specific characteristics. For example, biomedical CQA involves more complex interactive information between askers and answerers, while CQA techniques designed for the general field can only deal with single interactions between questions and candidate answers within a similar topic. To address the problem, we propose a novel neural network model for biomedical CQA. Our model adopts the bidirectional capsule network to focus on different aspects of biomedical questions and candidate answers, and merges high-level vector representations of questions and answers to capture abundant semantic information. Furthermore, to capture the meaning of Chinese characters, we incorporate the radical of Chinese characters embedding as auxiliary information to improve the performance of Chinese biomedical CQA. We conduct extensive experiments, and demonstrate that our model achieves significant improvement on the performance of answer selection in the Chinese biomedical CQA task.