[4O1-IS-2a] Machine learning

9:45 AM - 10:00 AM JST(12:45 AM - 1:00 AM UTC)

[4O1-IS-2a-04] Heterogeneous Treatment Effects of Push Notification Strategies on Streaming Platform Engagement: A Causal Forest Approach.

〇Aleksandr Yamamoto¹, Sato Shunichi¹, Asai Shuto¹, Takahiro Hoshino^1,2 (1. Keio University, 2. RIKEN Center for Advanced Intelligence Project (AIP))

regular

Keywords:

Machine Learning,Push Notifications,Streaming Platform,Heterogeneous Treatment Effects

Mobile applications increasingly rely on push notification to drive their user engagement, yet optimisation of notification timing tends to be heuristic. Current approaches are often uniform timing rules or simple A/B tests, overlooking individual-level behavioural heterogeneity in notification responsiveness. Moreover, excessive notification can lead to "notification fatigue", resulting in user disengagement or app uninstallation. To address these challenges, our study introduces a causal machine learning framework for personalised notification timing, based on the data from the randomised experimental data from a streaming platform. Our experimental design incorporates a 3-day interval between notifications to mitigate the notification fatigue effects. By incorporating variables such as time-of-day, day-of-week patterns and minute level timing offsets (+0/ +15/ +30), we estimate heterogeneous treatment effects to identify potential differential responses across user segments and notification schedules. This research advances both the methodological application of Causal Forest to mobile engagement optimisation and provides practical insights for streaming services seeking to personalise user communication while mitigating notification fatigue.