Carnegie Mellon University

Glance: Improving Click-through Rate on Mobile Lock Screen Advertisements

Capstone Team: Hugo Camou, Johnathan Khodr, Jamie Khoo, Andrew Lee

Glance delivers its users interactive content (cards) to their phone’s lock screens. “Clicks” on sponsored cards is a measure of success for a campaign. Dividing clicks by the number of impressions gives click-through rate (CTR), a key metric when comparing success across campaigns.

Glance already utilizes a sophisticated recommendation system to distribute cards to its users. (Oli et al., 2020) However, to maximize revenue, it is important to understand what motivates a user to interact with a sponsored card. Thus, our goal was to create a modeling methodology that improves upon the existing recommendation system, makes consistent predictions across multiple card types, and offers interpretable results.

By predicting clicks on a target glance at a user level, we were able to develop a modeling methodology that achieves each of these goals. We feature engineered user data and past click history to make our predictions and found that past click history was most predictive of future clicks.

Capstone Presentation