Use of clustering for consideration set modeling in recommender systems

Abstract

The cold-start problem has become a significant challenge in recommender systems. To solve this problem, most approaches use various user-side data and combine them with item-side information in their systems design. However, when such user data is not available, those methods become unfeasible. We provide a novel recommender system design approach which is based on two-stage decision heuristics. By utilizing only the item-side characteristics we first identify the structure of the final choice set and then generate it using stochastic and deterministic approaches.