A Reinforcement Learning and Synthetic Data Approach to Mobile Notification Management

Sutton, Rowan; Fraser, Kieran; Conlan, Owen

doi:10.1145/3365921.3365932

Cited by 4 publications

(1 citation statement)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using the notification text as input and the notification action as ground-truth, a RL agent was trained to predict whether a participant would open or dismiss a notification based on its textual content. A Deep Q-learning (DQN) agent was chosen as it has shown previously to be effective at managing push-notifications on behalf of a notification subscriber [14].…”

Section: Notification Deliverymentioning

confidence: 99%

Enticing notification text & the impact on engagement

Fraser

Conlan

2020

Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of The

Self Cite

View full text Add to dashboard Cite

Push-notifications are a design tool used by mobile and web apps to alert subscribers to new information. In recent years, due to widespread adoption of the technology and the shrinking level of user attention available, marketing techniques have been deployed to persuade subscribers to engage positively with notifications. One such technique, known as the curiosity gap, exploits Lowenstein's Information-Gap theory. This paper explores the impact of enticing notification text, instilled by the curiosity gap, on subsequent engagement actions. A classifier was defined to identify enticing language in notifications. Features commonly paired with enticing text were identified. Intelligent notification delivery agents, trained using data captured in-the-wild, were evaluated using enticing and non-enticing notifications to demonstrate the influence of enticing text. Additionally, a solution was proposed and briefly evaluated for limiting subscriber susceptibility to enticing notifications. CCS CONCEPTS • Human-centered computing → Smartphones; • Computing methodologies → Natural language processing; Reinforcement learning.

show abstract

Section: Notification Deliverymentioning

confidence: 99%