Content Card Analysis to Maximize User Engagement for DraftKings Sportsbook

Radcliffe, Owen; Bhagavathula, Nagavagmi; Schuman, Eric; Albuquerque, Abigail; Lavoie, Kristin

Student Work

Content Card Analysis to Maximize User Engagement for DraftKings Sportsbook

Public Deposited

DraftKings presents unique betting opportunities to its customers through highly visible "content cards'' in its app and on its website. A collection of content cards forms a "carousel'' that users can swipe through, select, and place a bet. When a new content card is added to a carousel, DraftKings must determine the position of the new card. This project revolves around predicting the optimal position for inserting a content card within a carousel to maximize engagement, e.g., bets per minute, for DraftKings. We experimented with a wide array of models to achieve this objective. Following thorough testing, we identified three models based on their performance: Random Forest, Gradient Booster, and Multilayer Perceptron. The training and testing processes for these models are highlighted extensively in this paper. The models utilized a number of hand-engineered features, taking into account various aspects such as player and team statistics, odds, legs and other relevant attributes of both incoming cards and those within the carousel. The models were trained and tested on NBA, NFL and MLB zones. With the models, we were able to predict engagement change at a higher accuracy than the baseline. We also constructed a model wrapper allowing for real-time use of the models. To further improve model accuracy, we suggest the addition of more data (including previous seasons for each sport) and techniques such as A/B testing of different carousel orders.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

Publisher