Mobile Vertical Ranking based on Preference Graphs
Yuta Kadotami(Waseda Univ.), Yasuaki Yoshida, Sumio Fujita, Tetsuya Sakai(Waseda Univ.)
ICTIR 2017, 2017/10
Information Retrieval Machine Learning Data Science
- We consider the problem of ranking relevant verticals for a given mobile search query so as to satisfy the average user. To this end, we utilise real mobile search click logs, and apply a graph contruction algorithm proposed by Agrawal et al. who tackled the problem of automatically assigning relevance labels to URLs for general web search. While Agrawal et al. ordered URLs based on pairwise preferences and then partitioned the ordered URL list to determine absolute relevance grades, our objective is to rank a given set of verticals for a given query, to help search engine companies select which verticals to include in a search engine result page for a small smartphone screen. We show that “Click Skip Other” preference rules consistently outperform more conservative rules such as “Click Skip Previous,” and that our best graph-based vertical ranking methods substantially and statistically significantly outperform a competitive baseline that ranks verticals based on click counts.
Mobile Vertical Ranking based on Preference Graphs（External Site Link）