As the annotation of multimedia documents uses multiple descriptors, it is possible to define multiple, semantically meaningful, similarity (or dissimilarity) relations among them. Therefore, for cases such as the mining of user interests for consumption of multimedia documents, based on usage history, where the clustering of documents is necessary, it is important to develop context aware clustering algorithms that are able to handle this type of information. In this paper we explain the relation between context, user interest and the multiple relations; furthermore, we present a clustering algorithm that is able to mine user interests from multi-relational data sets.