Patient preference analysis for online consultation based on user-generated content.

Abstract

Tailoring doctor recommendations to patients' needs and preferences is the core of online consultation platforms. Existing studies encountered challenges in capturing patients' personalized preferences towards different attributes of doctors in doctor selection, owing to the sparsity of individuals' medical consultation records. This study proposes a patient preference disaggregation analysis method to learn the preference models of patients from user-generated content and generate personalized doctor recommendations. A compensatory value function is employed to represent the preference model of a patient towards different attributes of doctors, indicating the compensatory mechanisms among these attributes. An optimization model is established to learn parameters in the value function from historical decision examples, incorporating regularization to prevent overfitting. These examples consist of doctor attribute performance derived from sentiment analysis of online reviews, factoring in reliability and popularity based on comprehensive indicators like diagnosis volume and ratings. Patients are segmented into cohorts (e.g., male, female, minor, severe cases), with preference models developed for each. An individual-cohort matching model estimates personalized preferences for tailored recommendations. Validation on the Dxy.com platform confirms that different patient cohorts have distinct attribute preferences, and our method enhances the patient medical experience through personalized recommendations.