Making money out of innovative communities on the Internet

Abstract

Online advertising has been around for more than a decade, and today most Internet users are familiar with the contextual ad targeting in online search, of which Google AdWords is a dominant example. The pay per click model (PPC) introduced by Goto.com (later Overture) in 1998 has been the main revenue model since Google adopted it in 2002, but the advertising industry is under increasing pressure to find a better alternative model as the conversion rate for ad clicks to actual purchase is very low. There is a huge potential for improving the experience as seen from both, advertiser, publisher and end user point of view. Enter the cost per install model where advertising cost is directly tied to the response of potential customers. While this reduces risk for advertiser spending, it demands precise customer targeting to create added value for the end user and secure revenue for the publisher. The authors present a system for content-based targeting where the user behaviour within Zedge’s massive mobile community is the core signal for segmenting users into preference groups that are used for targeted advertisements. Large scale clustering using Hadoop and machine learning is combined to optimize the performance.