Optimizing Forecasted Activity Notifications with Reinforcement Learning.

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Behavioral Science

Background:

• Optimizing notification timing is crucial for user engagement and minimizing disruptions.
• Forecasting daily activities and their probabilities presents challenges for effective reminder systems.
• Balancing task completion with user experience requires intelligent notification strategies.

Purpose of the Study:

• To propose a notification optimization method considering probabilistic forecasts of activities.
• To investigate the impact of low activity probabilities on notification effectiveness.
• To enhance user self-improvement by managing task completion and additional tasks.

Main Methods:

• Developed a notification optimization method using reinforcement learning.
• Incorporated probabilistic considerations for forecasted activities.
• Evaluated the method using existing datasets and newly collected field data from six participants (23 activities).

Main Results:

• The proposed method significantly improved notification response rates by up to 27.15% compared to baseline methods.
• Incorporating activity probability into the notification system enhanced user response rate and other performance criteria.
• Demonstrated a more effective approach to optimizing notifications by balancing usefulness and interruption.

Conclusions:

• Probabilistic considerations in activity forecasting lead to optimized notification strategies.
• Reinforcement learning is effective for balancing reminder utility and user interruption.
• The developed method offers a promising solution for intelligent notification systems in daily task management.