A personalized health recommendation system based on smartphone calendar events

Abstract

Many e-health services are available to users today, but they often suffer from lack of personalization. In this paper, we present a system to generate personalized health recommendations from various providers, based on classification of health related calendar events on the user’s smartphone. Due to privacy constraints, such personal data often cannot be uploaded to external servers, hence the classification and personalization has to run on the client device. We use a server to train our model to classify calendar events using SVM and fastText, while the prediction is run on the client device using the trained model. The class labels from the classified calendar events, weighted in order of recency, are used to build a vector, which we treat as a representation of user interest while personalizing the recommendations. This vector is used to re-rank health related recommendations obtained from third party providers based on relevance. We describe the implementation details of our system and some tests on its accuracy and relevance to provide relevant health related recommendations. While we used the calendar app to classify events, our system can also be extended for other apps such as messaging.