Comparing Click Logs and Editorial Labels for Training Query Rewriting
暂无分享,去创建一个
Clicks on web advertisements in response to web search queries is a major source of revenue for search companies. Query rewrites can signican tly increase the coverage of web advertisements. In previous work we focused on optimizing the relevance between the query issued by the web searcher, and rewritten queries used to place advertisements. In this preliminary study, we examine some features of query rewrites which are predictive of click-throughs on sponsored search listings retrieved for those rewrites, by mining web search-click logs. We also compare the features which are predictive of relevance (judged by human editors) and the clicks in user query logs during query rewriting. Our preliminary results suggest that similar features are predictive, and so we may be able to train our models on click log data in place of human relevance judgments.
[1] Susan T. Dumais,et al. Learning user interaction models for predicting web search result preferences , 2006, SIGIR.
[2] Carlos A. Hurtado,et al. A Statistical Model of Query Log Generation , 2006, SPIRE.
[3] Thorsten Joachims,et al. Accurately Interpreting Clickthrough Data as Implicit Feedback , 2017 .
[4] Benjamin Rey,et al. Generating query substitutions , 2006, WWW '06.
[5] Daniel C. Fain,et al. Predicting Click-Through Rate Using Keyword Clusters , 2006 .