Short Text Similarity Hybrid Algorithm for a Chinese Medical Intelligent Question Answering System

Abstract

The medical-oriented intelligent question answering (QA) system, which provides users with fast and accurate answers, has gradually caught the attention of the medical and health research community. In a QA system, one of the critical processes is determining how to calculate the textual similarity and match it, for example, the similarity between the question asked by a user and the question as it exists in the system template. However, the question texts are shorter, individual noisy words pose new challenges for semantic parsing of the entire text. In this paper, we propose a new shared layer-based convolutional neural network (SH-CNN) model to calculate the semantic similarity of Chinese short text. The SH-CNN uses a shared layer to extract the prominent features from a pair of questions and then yields textual similarity based on their features. Concurrently, our approach employs the term frequency-inverse document frequency (TF-IDF) algorithm in the feature extraction process, reducing the interference of noisy words during short text similarity calculations. The experimental results show that the textual similarity hybrid algorithm combining SH-CNN and TF-IDF achieves a successful performance in our intelligent QA system and demonstrates a meaningful application value.