User　analysis　is　an　important　part　of　social　network　analysis.　Most　existing　studies　model　users　separately　using　either　user-generated　contents　or　social　links　among　users.　In　this　paper　we　propose　to　model　users　on　the　Content　Curation　Social　Network　(CCSN)　in　a　unified　framework　by　mining　User-generated　contents　as　well　as　social　links.　We　propose　a　latent　Bayesian　model　Multi-level　LDA　(MLLDA)　that　represents　users　with　latent　user　interests　discovered　from　user-contributed　textual　description　and　social　links　formed　by　information　sharing.　We　demonstrate　that　MLLDA　can　produce　accurate　user　models　for　community　discovery　and　recommendation　on　the　CCSN.