Multi-level Composite Neural Networks for Medical Question Answer Matching

The online medical question answering community where patients can ask medical questions becomes quite popular in recent years. Deep learning techniques have been widely used in medical care field and a large number of Natural Language Processing tasks, which makes it possible to answer the medical question automatically. The challenge in extracting high-level semantic information from the question and the answer is the key issue in question answer matching. This paper proposes a multi-level composite convolutional neural networks framework to alleviate the issue in question answer matching. The model does not just stack multiple convolutional neural networks together, but extracts information from each layer and then concatenates the features at the end of the framework. As a consequence, the framework is able to better capture high-level semantics and reach a new state-of-the-art performance on cMedQA dataset.