Cross-Market Model Adaptation with Pairwise Preference Data for Web Search Ranking

摘要

Machine-learned ranking techniques au-tomatically learn a complex document ranking function given training data. These techniques have demonstrated the effectiveness and flexibility required of a commercial web search. However, man-ually labeled training data (with multiple absolute grades) has become the bottle-neck for training a quality ranking func-tion, particularly for a new domain. In this paper, we explore the adaptation of machine-learned ranking models across a set of geographically diverse markets with the market-specific pairwise prefer-ence data, which can be easily obtained from clickthrough logs. We propose a novel adaptation algorithm, Pairwise-Trada, which is able to adapt ranking models that are trained with multi-grade labeled training data to the target mar-ket using the target-market-specific pair-wise preference data. We present results demonstrating the efficacy of our tech-nique on a set of commercial search en-gine data.