This paper discusses the 3A recommender system that targets CSCL (computer-supported collaborative learning) and CSCW (computer-supported collaborative work) environments. The proposed system models user interactions in a heterogeneous graph. Then, it applies a personalized, contextual, and multi-relational ranking algorithm to simultaneously rank actors, activity spaces, and assets. The results of an empirical evaluation carried out on an Epinions dataset indicate that the proposed recommendation approach exploiting the trust and authorship networks performs better than user-based collaborative filtering in terms of recall