This document summarizes a research paper that proposes using orthogonal nonnegative matrix tri-factorization (ONMTF) to fuse model-based and memory-based collaborative filtering approaches. ONMTF is used to co-cluster users and items to obtain centroids that are then used to select similar users and items for predicting unknown ratings. Experimental results on movie rating datasets show the ONMTF approach improves prediction accuracy over other collaborative filtering methods.