A Hierarchical Model for Value Estimation in Sponsored Search

Sponsored search is a form of online advertising where adver- tisers bid for placement next to search engine results for spe- cic keywords. As search engines compete for the growing shares of online ad spend, it becomes important for them to understand what keywords advertisers value most, and what characteristics of keywords drive value. In this paper we pro- pose an approach to keyword value prediction that proceeds in two steps. We rst estimate values on high-volume key- words based on advertiser bids, assuming rational bidding behavior. We then t a hierarchical model on top of these estimates, drawing on demographic and textual features of keywords and taking advantage of the hierarchical struc- ture of sponsored search accounts. The predictive quality of our model is evaluated on fty high-spending advertising ac- counts on a major search engine. In the process of tting our model we uncover evidence that advertiser utility is additive across keywords, an implicit assumption in the literature to date. Our evaluation shows that our model outperforms several baselines for value inference, and that improvements are even more pronounced for large accounts. Besides the general evaluation of advertiser welfare, our approach has potential applications to keyword and bid suggestion.