This paper presents a novel way of building the user profile of concept network for personalized search. The user profile is defined as a concept network, in which each concept is approximately represented with the formal concept analysis (FCA) theory. We assume that a concept, called `session interest concept', subsume a user's query intention during a query session and it can reflect the user's preference. Whenever a user issues his/her query, a session interest concept is generated. Then, new concepts are merged into the current concept network (i.e., a user profile) in which recent user preferences are accumulated. According to FCA, a session interest concept is defined as a pair of extent and intent where the extent covers a set of documents selected by the user among the search results and the intent covers a set of keyword features extracted from the selected documents. And, in order to make a concept network grow, we need to calculate the similarity between a new concept and existing concepts, and to this end, we use a reference concept hierarchy called Open Directory Project. The user profile of concept network is eventually used to expand a user's initial query. The empirical results show that our approach improves the accuracy of search results in terms of personal preference.