Optimization of Just-in-Time Adaptive Interventions Using Reinforcement Learning

Momentary context data is an important source for intelligent decision making towards personalization of mobile phone notifications. We propose a reinforcement learning based personalized notification delivery algorithm, reasoning over momentary context data. Beyond the state of the art, we propose new approaches for faster convergence of the algorithm and jump start of learning performance at the beginning of the learning process. We test our approach in both simulated and real settings trying to optimize the timing of the notifications. Our eventual, practical aim is to make office workers more physically active during the work time. We compare the results obtained for standard and improved algorithms in both testbeds where improved versions yield better results.