A　recommender　system　aims　at　recommending　items　that　users　might　be　interested　in.　With　an　increasing　popularity　of　social　tagging　systems,　it　becomes　urgent　to　model　recommendations　on　users,　items,　and　tags　in　a　unified　way.　In　this　paper,　we　propose　a　framework　for　studying　recommender　systems　by　modeling　user　preferences　as　a　relation　on　(user,　item,　tag)　triples.　We　discuss　tag-aware　recommender　systems　from　two　aspects.　On　the　one　hand,　we　compute　associations　between　users　and　items　related　to　tags　by　using　an　adaptive　method　and　recommend　tags　to　users　or　predict　item　properties　for　users.　On　the　other　hand,　by　taking　the　similarity-based　recommendation　as　a　case　study,　we　discuss　similarity　measures　from　both　qualitative　and　quantitative　perspectives　and　k-nearest　neighbors　and　reverse　k-nearest　neighbors　for　recommendations.