A Topic Modeling Approach to Ranking

We propose a topic modeling approach to the prediction of preferences in pairwise compar-isons. We develop a new generative model for pairwise comparisons that accounts for multi-ple shared latent rankings that are prevalent in a population of users. This new model also captures inconsistent user behavior in a natural way. We show how the estima-tion of latent rankings in the new generative model can be formally reduced to the esti-mation of topics in a statistically equivalent topic modeling problem. We leverage recent advances in the topic modeling literature to develop an algorithm that can learn shared latent rankings with provable consistency as well as sample and computational complex-ity guarantees. We demonstrate that the new approach is empirically competitive with the current state-of-the-art approaches in pre-dicting preferences on some semi-synthetic and real world datasets.