Carolina Ballroom
Sponsored by Bank of America
Optimizing TV Ad Buys with Individual-level Viewership Models (303458)
*April Chen, Civis Analytics"Civis Media Optimizer is a web-based application that brings the targeting precision of digital to the reach of TV. We achieve this by building individual-level models that predict an organization’s target audience and their television viewership patterns. Then, using these data, Civis Media Optimizer picks the exact mix of media buys that maximizes an organization’s reach.
In this speed session I will focus on one important component of the Civis Media Optimizer – the model of television viewership patterns. I will demonstrate how I used historical viewership patterns, demographic information, seasonality trends, and clusters of show features to predict the likelihood that any individual would watch any network at any time. A few of the methodologies that will be covered include: handling large and sparse datasets, modeling for imbalanced target outcomes, and the tradeoffs between using out-of-core algorithms on a complete dataset and using in-core algorithms on a subsample of data. Finally, I will demonstrate the impact that this model and Civis Media Optimizer has on an organization’s outreach efforts."