A Reinforcement Learning and Synthetic Data Approach to Mobile Notification Management

Mobile push-notifications are the primary mechanism for communicating new information to smartphone users, however they can also have a negative impact on user emotions, reduce work effectiveness and decrease current task performance. Through analysing state-of-the-art research on mobile Notification Management Systems, it was identified that few open-source notification data sets and, corresponding benchmarks, have been created and the majority of NMSs apply supervised learning methods. This paper investigates the use of a, freely shareable, synthetic mobile notification data set for developing and evaluating NMS performance using Reinforcement Learning. A Q-learning and Deep Q-learning agent were trained using synthetic data and an OpenAI Gym environment was created for evaluation. Final results illustrated that the Q-learning and Deep Q-learning agents could predict a users action toward notifications with ≈80% success when trained and evaluated upon real or synthetic data and ≈65% success when trained on synthetic and evaluated upon real notification data.