We consider the problem of generating interpretable recommendations byidentifying overlapping co-clusters of clients and products, based only onpositive or implicit feedback. Our approach is applicable on very largedatasets because it exhibits almost linear complexity in the input examples andthe number of co-clusters. We show, both on real industrial data and onpublicly available datasets, that the recommendation accuracy of our algorithmis competitive to that of state-of-art matrix factorization techniques. Inaddition, our technique has the advantage of offering recommendations that aretextually and visually interpretable. Finally, we examine how to implement ourtechnique efficiently on Graphical Processing Units (GPUs).
展开▼
