A Multi-granularity Fusion Neural Network Model for Medical Question Classification

Knowledge-Based Question Answering (KBQA) is a novel method for Question Answering. To construct the semantic parser for a given question, it is vital to effectively encode the existing question for question classification. In this work, we propose a novel Multi-granularity fusion deep learning architecture that consists of sequence encoding, phrase vector recombination and feature extraction for the given question strings. We adopt Bi-GRU to learn features by different forms for question classification. In addition, attention mechanism is incorporated in the proposed model. We construct the local question answering base on clinical neurologic. We deploy plenty of comparision experiments among our proposed multi-granularity fusion model and other well-known methods. Experiments show that our proposed method achieves the highest accuracy.