Collaborative tagging, the process of assigning tags to shared content by many users, has emerged as an important way to share resources on the Web. Tags are freely chosen keywords. They not only help the users to find and organize online resources, but also provide meaningful indexing which can be used by other research areas. In this thesis, we focus on the problem of finding similar users by exploiting tagging data. We propose user profiling method by tags which are attached on user``s bookmarked item instead of direct tag usage. Also we consider the semantic in tags by applying Latent Semantic Analysis in tag-user matrix. Compared with user profile in recommender system, our proposed method only use tagging data and does not need rating or other additional data. We conduct two kinds of evaluations on the data from CiteULike. First one uses group membership data for answer set and second one uses Watchlist data as answer set. The experimental result shows that our suggested method improves recall, precision and F-measure with various situations. The improvement by each method is not consistent in various situations, but the combination of both proposed methods outperforms in almost all situations.