This paper introduces new Web usage mining tools designed to help characterizing user accesses on websites. Our approach relies on a categorization of Web user sessions to help identifying and understanding major trends in the navigations. The novelty of our work mainly relies in the clustering method that is a fast unsupervised ant-inspired clustering algorithm paired with new similarity measures that handle sessions either as a sequence or an unordered set of Web pages. Our algorithm is evaluated on real Web log files of a French museum that contains more than 39000 user sessions over one month and a half. Our experiments show that our algorithm can build meaningful sessions clusters that can help infer Web users' motivations.