Collaborative filtering systems (CFs) aim to provide relevant items for users on the web. Most of existing CFs are based on matrix factorization and k nearest neighbors methods. Unfortunately both approaches are expensive in terms of computational time, and do not treat missing data in the user-item rating matrix. The computational time flaw, can be addressed by using co-clustering methods, which involve the user and item spaces simultaneously. However, the latter approaches still need an efficient strategy for handling missing values. In this work, we propose an effective method for handling unobserved ratings, allowing a better use of co-clustering approaches in CF. Furthermore we propose an interactive representation of coclustering results. Based on bipartite graphs, this representation allows an easy interpretation and sense making of the preferences between user and item clusters.