NCDREC: A Decomposability Inspired Framework for Top-N Recommendation

Building on the intuition behind Nearly Decomposable systems, we propose NCDREC, a top-N recommendation framework designed to exploit the innately hierarchical structure of the item space to alleviate Sparsity, and the limitations it imposes to the quality of recommendations. We decompose the item space to define blocks of closely related elements and we introduce corresponding indirect proximity components that try to fill in the gap left by the inherent sparsity of the data. We study the theoretical properties of the decomposition and we derive sufficient conditions that guarantee full item space coverage even in cold-start recommendation scenarios. A comprehensive set of experiments on the Movie Lens and the Yahoo!R2Music datasets, using several widely applied performance metrics, support our model´s theoretically predicted properties and verify that NCDREC outperforms several state-of-the-art algorithms, in terms of recommendation accuracy, diversity and sparseness insensitivity.