Ranking via Sinkhorn Propagation

It is of increasing importance to develop learning methods for ranking. In contrast to many learning objectives, however, the ranking problem presents difficulties due to the fact that the space of permutations is not smooth. In this paper, we examine the class of rank-linear objective functions, which includes popular metrics such as precision and discounted cumulative gain. In particular, we observe that expectations of these gains are completely characterized by the marginals of the corresponding distribution over permutation matrices. Thus, the expectations of rank-linear objectives can always be described through locations in the Birkhoff polytope, i.e., doubly-stochastic matrices (DSMs). We propose a technique for learning DSM-based ranking functions using an iterative projection operator known as Sinkhorn normalization. Gradients of this operator can be computed via backpropagation, resulting in an algorithm we call Sinkhorn propagation, or SinkProp. This approach can be combined with a wide range of gradient-based approaches to rank learning. We demonstrate the utility of SinkProp on several information retrieval data sets.

[1]  Richard Sinkhorn,et al.  Concerning nonnegative matrices and doubly stochastic matrices , 1967 .

[2]  T. Raghavan,et al.  Nonnegative Matrices and Applications , 1997 .

[3]  Philip A. Knight,et al.  The Sinkhorn-Knopp Algorithm: Convergence and Applications , 2008, SIAM J. Matrix Anal. Appl..

[4]  Todd K. Moon,et al.  Sinkhorn Solves Sudoku , 2009, IEEE Transactions on Information Theory.

[5]  Geoffrey E. Hinton,et al.  Deep Belief Networks for phone recognition , 2009 .

[6]  Alexander J. Smola,et al.  Direct Optimization of Ranking Measures , 2007, ArXiv.

[7]  Honglak Lee,et al.  Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations , 2009, ICML '09.

[8]  Mingrui Wu,et al.  Gradient descent optimization of smoothed information retrieval metrics , 2010, Information Retrieval.

[9]  H. Balakrishnan,et al.  Polynomial approximation algorithms for belief matrix maintenance in identity management , 2004, 2004 43rd IEEE Conference on Decision and Control (CDC) (IEEE Cat. No.04CH37601).

[10]  Tao Qin,et al.  LETOR: A benchmark collection for research on learning to rank for information retrieval , 2010, Information Retrieval.

[11]  J. Munkres ALGORITHMS FOR THE ASSIGNMENT AND TRANSIORTATION tROBLEMS* , 1957 .

[12]  Mark Huber,et al.  Fast approximation of the permanent for very dense problems , 2008, SODA '08.

[13]  Tie-Yan Liu,et al.  Learning to rank: from pairwise approach to listwise approach , 2007, ICML '07.

[14]  Manfred K. Warmuth,et al.  Learning Permutations with Exponential Weights , 2007, COLT.

[15]  Alexander Dekhtyar,et al.  Information Retrieval , 2018, Lecture Notes in Computer Science.

[16]  Maksims Volkovs,et al.  BoltzRank: learning to maximize expected ranking gain , 2009, ICML '09.

[17]  Richard Sinkhorn A Relationship Between Arbitrary Positive Matrices and Doubly Stochastic Matrices , 1964 .

[18]  Hang Li,et al.  AdaRank: a boosting algorithm for information retrieval , 2007, SIGIR.

[19]  Stephen E. Robertson,et al.  SoftRank: optimizing non-smooth rank metrics , 2008, WSDM '08.

[20]  Jaana Kekäläinen,et al.  Cumulated gain-based evaluation of IR techniques , 2002, TOIS.

[21]  Alistair Moffat,et al.  Rank-biased precision for measurement of retrieval effectiveness , 2008, TOIS.