Cluster-based Smoothing of Sparse Ranking Signals in Mobile Local Search

Users increasingly rely on their mobile devices to search for local entities, typically businesses, while on the go. Recent work has recognized unique ranking signals in mobile local search (e.g., distance, customer rating, and number of reviews), and has proposed various ways of leveraging these signals for ranking. However, these techniques have overlooked a major challenge that is amplified in the case of mobile local search: data sparseness. In this work, we exploit domain knowledge about businesses to cluster them based on either the category of the business or the parent chain store that the business belongs to. We then smooth individual business’ sparse ranking signals based on the hypothesis that businesses in the same cluster share similar ranking signals. Our experimental evaluation using 14 months of real mobile local search logs, shows that the proposed cluster-based smoothing of these ranking signals can improve mean average precision by 5%.