OClustR: A new graph-based algorithm for overlapping clustering

• Published in: Neurocomputing (Amsterdam) Vol. 121; pp. 234 - 247
• Main Authors: Pérez-Suárez, Airel, Martínez-Trinidad, José F., Carrasco-Ochoa, Jesús A., Medina-Pagola, José E.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 09.12.2013; Elsevier
• Subjects: Algorithmics. Computability. Computer arithmetics; Algorithms; Applied sciences; Clustering; Clusters; Collection; Computer science; control theory; systems; Computer systems and distributed systems. User interface; Construction; Data mining; Data processing. List processing. Character string processing; Exact sciences and technology; Filtering; Graph-based algorithms; Information retrieval. Graph; Memory organisation. Data processing; Overlapping clustering; Software; Strategy; Theoretical computing; Tracking; Data mining; Overlapping clustering; Graph-based algorithms; Cluster analysis; Data analysis; Filtering; Electronic document; Cluster; Information retrieval; Social network; Graph theory; Overlap; Digital information; Graph connectivity; News; Categorization
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2013.04.025

Clustering is a Data Mining technique, which has been widely used in many practical applications. From these applications, there are some, like social network analysis, topic detection and tracking, information retrieval, categorization of digital libraries, among others, where objects may belong to more than one cluster; however, most clustering algorithms build disjoint clusters. In this work, we introduce OClustR, a new graph-based clustering algorithm for building overlapping clusters. The proposed algorithm introduces a new graph-covering strategy and a new filtering strategy, which together allow to build overlapping clusterings more accurately than those built by previous algorithms. The experimental evaluation, conducted over several standard collections, showed that our proposed algorithm builds less clusters than those built by the previous related algorithms. Additionally, OClustR builds clusters with overlapping closer to the real overlapping in the collections than the overlapping generated by other clustering algorithms. •A new overlapping clustering algorithm, called OClustR, is proposed.•OClustR introduces a new graph-covering strategy and a new filtering strategy.•OClustR outperforms state-of-the-art algorithms in several collections.•OClustR solves the limitations of the state-of-the-art algorithms.