The objective of our research is to create a universal tool for recommending non-visited interesting web pages as well as experts working in the same field of specialty. We accentuate practical adaptability of user profiles. User profiles are generated on the basis of Suffix Tree Clustering (STC) algorithm, which is similar to creating an inverted list of phrases occurring in a document collection. We are computing similarity of characteristic phrases identified by STC in order to find clusters of phrases. Phrases linked by similarity relationships form a phrase association graph. Clusters of phrases generated by our tool define interests of each user. We have tested the system by means of various document collections, such as Reuters Corpus Volume One – RCV1, 20Newsgroups, CTK – Czech Press Agency and Reuters-21578. Experimental results based on our extensive simulations as well as real-life environment are presented in the paper. Precision of our recommender system is 85 to 95 %.