CareGraph: A Graph-based Recommender System for Diabetes Self-Care

Abstract

In this work, we build a knowledge graph that captures key attributes of content and notifications in a digital health platform for diabetes management. We propose a Deep Neural Network-based recommender that uses the knowledge graph embeddings to recommend health nudges for maximizing engagement by combating the cold-start and sparsity problems. We use a leave-one-out approach to evaluate the model. We compare the proposed model performance with a text similarity and Deep-and-Cross Network-based approach as the baseline. The overall improvement in Click-Through-Rate prediction AUC for the Knowledge-Graph-based model was 11%. We also observe that our model improved the average AUC by 5% in cold-start situations.

Tasks

Reproductions