Improving cold-start recommendations using item-based stereotypes

Abstract Recommender systems (RSs) have become key components driving the success of e-commerce and other platforms where revenue customer satisfaction is dependent on user’s ability to discover desirable items in large catalogues. As number users a platform grows, computational complexity sparsity problem constitute important challenges for any recommendation algorithm. In addition, most widely studied filtering-based RSs, while effective providing suggestions established items, are known their poor performance new user item (cold-start) problems. Stereotypical modelling promising approach solving these A stereotype represents an aggregation characteristics or which can be used create general classes. We propose set methodologies automatic generation stereotypes address cold-start problem. The novelty proposed rests findings that built independently user-to-item ratings improve both metrics during phases. resulting RS with machine learning algorithm as solver, improved gains due rate-agnostic orthogonal obtained using more sophisticated solvers. paper describes how such item-based evaluated via series statistical tests prior being recommendation. improves quality under variety significantly reduces dimension model.