Multidimensional data are getting increasing attentionudfrom researchers for creating better recommender systemsudin recent years. Additional metadata provides algorithmsudwith more details for better understanding the interactionudbetween users and items. While neighbourhood-basedudCollaborative Filtering (CF) approaches and latent factorudmodels tackle this task in various ways effectively, theyudonly utilize different partial structures of data. In thisudpaper, we seek to delve into different types of relations inuddata and to understand the interaction between users anduditems more holistically. We propose a genericudmultidimensional CF fusion approach for top-N itemudrecommendations. The proposed approach is capable ofudincorporating not only localized relations of user-user anduditem-item but also latent interaction between alluddimensions of the data. Experimental results showudsignificant improvements by the proposed approach inudterms of recommendation accuracy.
展开▼
