Multi-Scale Attentive Interaction Networks for Chinese Medical Question Answer Selection

The past few years have witnessed a trend that the deep learning techniques have been increasingly applied in healthcare due to the explosive growth of big data. The online medical community, where users can ask qualified doctors about medical questions with just a few keystrokes and mouse clicks anytime and anywhere, has become quite popular recently. In this paper, we investigate the problem of Chinese medical question answer selection, which is a crucial subtask of automatic question answering and fairly challenging because of its language and domain characteristics. We introduce an end-to-end multiscale interactive networks framework to address the issue. The framework consists of several multi-scale deep neural layers which extract the deep semantic information of medical text from different levels of granularity, shortcut connections which prevent network degradation problem, and attentive interaction which mines the correlation between questions and answers. To evaluate our framework, we update and expand a dataset called cMedQA v2.0. Experimental results demonstrate that our model outperforms the existing state-ofthe-art models with noticeable margins.