Topic
In recent years online advertising has become increasingly ubiquitous and effective. Advertisements shown to visitors fund sites and apps that publish digital content, manage social networks, and operate e-mail services. Given such large variety of internet resources, determining an appropriate type of advertising for a given platform has become critical to financial success.
In this talk I will give an overview of different approaches we took to adapt ad targeting to a particular platform, including Yahoo Search, Tumblr, and Yahoo Email. Each of these platforms has an unique set of characteristics and signals and that can be leveraged to achieve successful ad targeting. For example, in web search, users communicate a very clear intent through a search query that allows effective ad targeting via bid term keywords defined by advertisers.
However, the search engine can help improve reach and quality of the advertising campaign by finding additional queries that deem relevant to the advertised product. For this purpose, we designed an algorithm that relies on search sessions, ad clicks as well as implicit negative signals, such as short dwell time and skipped ads, to learn latent representations of queries and ads that can be utilized for efficient sponsored search matching. On the other hand, designing an ad targeting platform for a social network such as Tumblr involves mapping the content of blogs and user actions, including likes, reblogs and follows, to an advertising taxonomy.
For this purpose we developed a semi-supervised framework that leverages intensity and recency of categorized user content and actions to create targeting audiences that have been put at advertisers’ disposal. Finally, our work on the next generation of native ad experience in Yahoo Email client focused on serving product recommendation ads based on users’ previously purchased products extracted from email receipts. In my talk, I will show both offline and online experimental results reported in our research papers. The results show that our approaches significantly outperform existing state-of-the-art, substantially improving a number of key business metrics.