Exploring graph-based global similarity estimates for quality recommendations
Deepa Anand, Kamal K. Bharadwaj · International Journal of Computational Science and Engineering · 2014
Data sparsity or the insufficiency of past user preferences in predicting future user needs continues to be a major challenge for RS engines. We propose a solution to the sparsity problem by exploring similarity measures that capture the global patterns of commonality between users or items by leveraging on indirect ways of connecting users (items) through a user (item) graph. Entities (users or items) sharing common features are connected to each other by edges weighted by their proximity or distance. Graph-based techniques, for estimating transitive similarity between entities not directly connected, are exploited to bring the entities closer thus facilitating collaboration. Furthermore, we also propose a combined user-item graph approach for exploiting the similarity between users preferring similar items (and vice versa). In this work, we have suggested alternatives to the already existing global similarity assessment and we aim to investigate the appropriateness of the proposed techniques under differing data features.