A deep learning model for online doctor rating prediction

Predicting doctor ratings is a critical task in the healthcare industry. A patient usually provides ratings to a few doctors only, leading to the data sparsity issue, which complicates the rating prediction task. The study attempts to improve the prediction methodologies used in the doctor rating prediction systems. The study proposes a novel deep learning (DL) model for online doctor rating prediction based on a hierarchical attention bidirectional long short-term memory (ODRP-HABiLSTM) network. A hierarchical self-attention bidirectional long short-term memory (HA-BiLSTM) network incorporates a textual review's word and sentence level information. A highway network is used to refine the representations learned by BiLSTM. The resulting latent patient and doctor representations are utilized to predict the online doctor ratings. Experimental findings based on real-world doctor reviews from across two medical specialties demonstrate the proposed model's superior performance over state-of-the-art benchmark models. In addition, robustness analysis is used to strengthen the findings.