Collaborative Clustering of XML Documents

This paper presents a distributed collaborative approach to XML document clustering. According to a previous study, XML documents are mapped to a transactional domain, based on a data representation model which exploits the notion of XML tree tuple. This XML transactional model is well-suited to the identification of semantically cohesive substructures from XML documents, according to structure as well as content information. The proposed clustering framework employs a centroid-based partitional clustering paradigm in a distributed environment. Each peer in the network is allowed to compute a local clustering solution over its own data, then exchanges cluster centroids with other peers. The exchanged centroids correspond to recommendations offered by a peer to peers allowed to compute global representatives. Exploiting these recommendations, each peer becomes responsible for computing a global set of centroids for a given set of clusters. The overall clustering solution is hence computed in a collaborative way according to data from all the peers. Our approach has been evaluated on real XML document collections varying the number of peers. Results have shown that collaborative clustering leads to accurate overall clustering solutions with a relatively low load in the network.