Latent Ranking Analysis Using Pairwise Comparisons
Younghoon Kim, Wooyeol Kim, Kyuseok Shim · 2014
Ranking objects is an essential problem in recommendation systems. Since comparing two objects is the simplest type of queries in order to measure the relevance of objects, the problem of aggregating pair wise comparisons to obtain a global ranking has been widely studied. In order to learn a ranking model, a training set of queries as well as their correct labels are supplied and a machine learning algorithm is used to find the appropriate parameters of the ranking model with respect to the labels. In this paper, we propose a probabilistic model for learning multiple latent rankings using pair wise comparisons. Our novel model can capture multiple hidden rankings underlying the pair wise comparisons. Based on the model, we develop an efficient inference algorithm to learn multiple latent rankings. The performance study with synthetic and real-life data sets confirms the effectiveness of our model and inference algorithm.