Predicting the quality of online health expert question-answering services with temporal features in a deep learning framework

Currently, online health expert question-answering (HQA) services are the most popular services for health consumers to search online health information due to their convenience and the cost-effectiveness. However, one main challenge of these services is that the quality of the answers posted by physicians varies over a large spectrum. Physicians may not have sufficient time to answer the questions or just want to post advertisements in the answers. To determine high-quality answers, we can apply classification models on them. However, it is critical to include the corresponding features in the classification models. To further enhance the model performance, we propose a set of novel temporal features based on the characteristics of HQA services. Moreover, we propose a deep learning framework, referred to as a collaborative decision convolutional neural network (CDCNN), to learn the semantic non-linear features embedded in the data. By exploiting the learned non-linear semantic features, we can apply the factorization machines (FM) to obtain the quality score of an answer. Finally, we conduct extensive experiments to show the advantages of the proposed framework.