AdScorer: an event-based system for near real-time impact analysis of television advertisements (industry article)

Abstract

The media measurement industry is in turmoil, with the old prediction-based models being challenged by more accurate measurement techniques, based on actual viewer behaviour drawn from much larger sample selections. As measurement methods converge across different types of media, the online/offline measurement divide will diminish. Television is one such medium that has traditionally required offline measurements. Advertisers are, for the most part, still accepting predictions and historical behaviors rather than current facts. Despite the limited accountability, yearly spendings on television advertisements are still much higher than for any other medium, and rising. In this paper, we present AdScorer, a scoring system for television advertisements. Our system is based on event stream processing techniques, and can compute scores for advertisements in near real-time based on channel change events from viewer set-top boxes. Our results show that AdScorer is capable of delivering detailed scores on a per-advertisement spot basis for a whole block of commercials, immediately after the commercial break has ended. The scores include regional breakdowns with viewer numbers and shares for each geographical region of Norway as well as national scores. Our evaluation of AdScorer demonstrates that it is capable of scoring numerous channels simultaneously. In our experiments, we used one machine to analyze five channels, but our system can easily scale to support hundres of channels by adding more machines.