Bhanu Chandra Vattikonda

Research

• Improving Ad Relevance
  I worked on improving relevance ranker for sponsored search using knowledge from organic results. Turns out most published work in this space has not tried to take advantage of the intent expressed by the advertiser through bidding keyword. Partly this is because bidding keyword is already present in the ad creative. We worked on using the bidding keyword and expanding on it using organic results for the bidding keyword. From this, we extracted features that are strong indicators of the relevance of an ad to a user query. With these extra features we achieved a 2.7% improvement in precision-recall AUC over existing production system.


• Advertising Effectiveness
  I conducted a large-scale measurement of the effectiveness of search ads-measured in terms of incremental conversion gains. We developed a simple metric called net acquisition benefit (NAB) that admits comparisons between the efficacy of different ad campaign strategies without access to advertisers' private financial information. We analyze about a months' worth of data from a stop search engine and presented case studies on three common campaign strategies used by advertisers using NAB: cannibalization, poaching, and ad extensions. We identify cases where these campaign strategies are justified. Advertisers and ad agencies can replicate our methodology to apply it to other strategies of interest.


• Affiliate Spam on Twitter
  We conducted a five-month study of affiliate marketing on Twitter, in an attempt to quantify both the prevalence and effectiveness of touting tweets. Surprisingly, we found that only 0.5% of tweeted URLs are part of an affiliate marketing scheme (compared to a previously reported 4.5% rate for spam), perhaps because the conversion rate would need to be shockingly high---on the order of 15% or more---for touting to be a sustainably profitable enterprise on Twitter.

Publications