Position-Aware ListMLE: A Sequential Learning Process for Ranking

Abstract

ListMLE is a state-of-the-art listwise learning-to-rank algorithm, which has been shown to work very well in application. It defines the probabil-ity distribution based on Plackett-Luce Model in a top-down style to take into account the position information. However, both empirical contradic-tion and theoretical results indicate that ListM-LE cannot well capture the position importance, which is a key factor in ranking. To amend the problem, this paper proposes a new listwise rank-ing method, called position-aware ListMLE (p-ListMLE for short). It views the ranking prob-lem as a sequential learning process, with each step learning a subset of parameters which maxi-mize the corresponding stepwise probability dis-tribution. To solve this sequential multi-objective optimization problem, we propose to use lin-ear scalarization strategy to transform it into a single-objective optimization problem, which is efficient for computation. Our theoretical s-tudy shows that p-ListMLE is better than ListM-LE in statistical consistency with respect to typi-cal ranking evaluation measure NDCG. Further-more, our experiments on benchmark datasets demonstrate that the proposed method can sig-nificantly improve the performance of ListMLE and outperform state-of-the-art listwise learning-to-rank algorithms as well. 1