Adapting Online Advertising Techniques to Television

Adapting Online Advertising Techniques to Television

Sundar Dorai-Raj (Google Inc., USA), Yannet Interian (Google Inc., USA), Igor Naverniouk (Google Inc., USA) and Dan Zigmond (Google Inc., USA)
Copyright: © 2011 |Pages: 18
DOI: 10.4018/978-1-60960-189-8.ch009

Abstract

The availability of precise data on TV ad consumption fundamentally changes this advertising medium, and allows many techniques developed for analyzing online ads to be adapted for TV. This chapter looks in particular at how results from the emerging field of online ad quality analysis can now be applied to TV.
Chapter Preview
Top

Introduction

As online advertising has exploded in the past decade, it has often been contrasted with traditional media such as television, print, and radio. The inherently connected nature of online content has enabled unprecedented tracking and analysis of online advertising, and the resulting explosion of data has allowed Internet companies to develop ever more sophisticated algorithms for allocating and pricing advertising inventory. Using user-initiated signals (like “clickthrough”), these companies can indirectly measure the relevance of ads in specific contexts, and build models to predict which ads will most interest future users and maximize revenues for online publishers (Richardson et al., 2007).

In contrast, traditional television measurement has typically relied on relatively small panels of pre-selected households to report their viewing behavior. This approach often has meant that reliable measurements took days or weeks to produce, and could not be produced at all for niche programming that appealed to very small audiences. These methods also did not generate enough data to build determine which TV ads were most appealing to viewers, nor to predict which ads would be most appealing in the future.

However, a new source of TV-related data has emerged in recent years, one that is closer to Internet scale. The set-top boxes (STBs) used by most cable and satellite TV subscribers are often capable of collecting data on viewing behavior. These data can then be stripped of personally-identifiable information and anonymously aggregated, allowing for very detailed measurement of television viewing behavior. While previous panels collected data from thousands or perhaps tens of thousands of households, set-top box data are available from many million US households and similar numbers in other countries.

With these data in hand, it is now possible for television to adopt many of the analytical techniques pioneered in online advertising. In particular, ads can be scored for relevance or quality based on statistical models that predict how viewers are likely to respond. Ad inventory can then be allocated so as to maximize relevance, and to compensate publishers for any loss of audience due to ads.

This chapter will explore this emerging discipline. We will introduce the basic statistical techniques underlying the approach, and give examples for how such models can be implemented in software. We will also discuss applications of these models to television advertising, and some of the issues raised in the application of these techniques.

TV Audience Measurement

Panel-based audience measurement has a long history. In the US, Arthur Nielsen began measuring television audiences in 1950 based on a nationwide sample of 300 households (Nielsen, 2009a). Because there were only 48 commercial television stations in the US at the time and no more than a handful of viewing choices in any one local area, the audience of any given station could be adequately estimated with such a small sample. The Nielsen Company continues measuring TV audiences today, now with a sample of over 9,000 households (Nielsen, 2009b). Understanding and using these audience ratings has evolved into a discipline unto itself (see, for example, Webster et al., 2006).

Even the modern, expanded panel, however, is at times unable to measure the increasing fragmented TV audience flocking to niche programming (Bachman, 2009). For television ads (as opposed to programs), this bias is further compounded: attempts to judge the reaction of TV audiences to ads have focused on only the most popular programming. For the 2009 Super Bowl, for example, Nielsen published a likeability score and a recall score for the top ads [1]. The scores were computed using 11,466 surveys, and Nielsen reported only on the top 5 best-liked ads and most-recalled ads.

Several companies have started using data from STBs to measure TV audiences. In addition to Google, TNS, CANOE, Retrak, Tivo, and The Nielsen Company itself are using STB data (Mandese, 2009). Several of these companies will make such data available to media researchers on a subscription basis.

Complete Chapter List

Search this Book:
Reset