Topic Judgment Helps Question Similarity Prediction in Medical FAQ Dialogue Systems

As a prominent application of artificial intelligence in healthcare, medical dialogue systems have a promising future. In this paper, we built a medical Frequently Asked Questions (FAQ) dialogue system using an XGBoost classifier based on handcrafted features. Moreover, we propose to improve accuracy in domain-specific information retrieval by adding a topic judgment module to general sentence similarity prediction. As the size of the labeled medical corpus in Chinese is very limited, we built a corpus of Chinese medical FAQ from mvyxws https://www.mvyxws.com/ and annotated a small part of it. We applied our topic judgment module to the XGBoost model, a general sentence similarity model, and Baidu short text similarity API. Testing on two datasets, the models can be improved by adding the topic judgment in most cases.