PROJECT TITLE :

Real-Time Learning from an Expert in Deep Recommendation Systems with Application to mHealth for Physical Exercises

ABSTRACT:

In today's increasingly digital world, recommendation systems are playing an increasingly crucial role. They have discovered applications in a variety of fields, including music platforms like Spotify and movie streaming services like Netflix, amongst others. A lesser amount of research effort has been put into developing recommendation systems for physical exercise. A sedentary lifestyle has become the primary cause of a number of diseases as well as an important factor in the cost of medical care. In this article, we develop a recommendation system with the goal of suggesting daily exercise activities to users based on the users' previous activity, profiles, and other users who are similar to them. The user-profile attention mechanism and the temporal attention mechanism are both incorporated into the developed recommendation system. This system makes use of a deep recurrent neural network. In addition, exercise recommendation systems are very different from streaming recommendation systems in the sense that we cannot collect click feedback from participants in exercise recommendation systems. This is a significant difference between the two types of systems. As a result, we suggest an active learning process that operates in real time and includes an expert. The active learner computes the degree of uncertainty posed by the recommendation system at each time step and for each user. When the level of certainty is deemed insufficient, the active learner consults an expert for advice. In this paper, we derive the probability distribution function of marginal distance, and we use it to determine when it is appropriate to consult with subject matter experts for input. After integrating a real-time active learner into the recommendation system, our experimental findings on a mHealth dataset as well as the MovieLens dataset demonstrate an improvement in accuracy.