The Business Next Door: Click-Through Rate Modeling for Local Search

Computational advertising has received a tremendous amount of attention from the business and academic community recently. Great advances have been made in modeling click-through rates in well studied settings, such as, sponsored search and context match. However, local search has received relatively little attention. Geographic nature of local search and associated local browsing leads to interesting research challenges and opportunities. We consider a novel application of relational regression to local search. The approach allows us to explicitly control and represent geographic and category-based neighborhood constraints on the samples that result in superior click-through rate estimates. Our model allows us to estimate an interpretable inherent ‘quality’ of a business listing, which reveals interesting latent information about listings that is useful for further analysis.