Using Dimensionality Reduction to Improve Similarity Judgements for Recommendation

Recommendation Explorer is an experimental recommender system that attempts to address the provisional and contextual nature of user information needs by coupling the system's interface and recommendation algorithms. This study reports on the development and evaluation of a new similarity module for RecEx. Based on dimensionality reduction via the Singular Value Decomposition (SVD), the new module discovers high-order relationships among database items, thus obtaining a robust model of item-item similarity.