A method for discovering the knowledge of item rank from consumer reviews

Abstract

When observing a collection of ranked items, we may be interested in the questions of why and how one item is ranked over another. This paper presents a method for discovering the knowledge about the rank of the items from consumer reviews. We formulate the questions of interest as a single biconvex minimization problem which has a relationship with SVM(Support Vector Machines). To facilitate the process of knowledge discovery, we propose a two-stage learning algorithm for discovering knowledge from small data. Finally, we evaluate the method by showing our simulation and experiment results.