Efficient Health Class Recommendations for Kaiser Permanente Members: A Scalable Embedding-Based Approach
Mohammad Amir Sharif, Phuong Hoang, Justin C. Huang, Wenshan Wang, David Molina, Ajay Kumar, Suhai Liu, Faizan Javed · 2024
Kaiser Permanente (KP) is dedicated to meeting its members’ healthcare information needs through digital innovations, including health classes and programs. With a vast array of available classes, navigating and finding relevant ones poses a significant challenge. To address this, we present a personalized health class recommendation system on kp.org, leveraging users’ online activity and health conditions. Our approach employs embedding-based representations to capture semantic and contextual information efficiently. However, computation on these large number of embeddings is often not efficient. We introduce a scalable and distributed method that integrates different health class metadata and user profiles, enhancing recommendation accuracy. Additionally, we introduce a rank-aggregation method to integrate diverse business metrics. Evaluation results demonstrate improved performance in both machine learning metrics and clinical expert reviews.