170 likes | 301 Vues
This work by L. Moussiades and A. Vakali presents an innovative graph-clustering algorithm tailored for densely interconnected networks. The proposed method, backed by experimental validation, notably enhances cluster connectivity. The framework addresses conventional clustering limitations by ensuring refined clusters exhibit stronger internal connections compared to external links. The study details the motivation behind this approach, objectives achieved, as well as a comprehensive methodology and experimental results involving artificial datasets. This algorithm shows promise for applications in hierarchical agglomerative clustering.
E N D
Clustering dense graphs: A web site graph paradigm Author :L. Moussiades, A. Vakali Presented : Fen-Rou Ciou IPM, 2010
Outlines • Motivation • Objectives • Methodology • Experiments • Conclusions • Comments
Motivation • A conventional cluster number of links connected a vertex to its cluster is higher than the number of links connected the vertex to the remaining graph.
Objectives • To propose a graph-clustering algorithm is proved a refined cluster are more strongly connected with their cluster than with any other cluster.
Methodology Max • Schematic diagram
Methodology • Basic definition and notations
Methodology • Basic definition and notations
Methodology • Criterion function ICR
Methodology • Algorithm AICR
Experiments Artificial Data
Experiments Purity for clustering solutions
Experiments • Amod on ds1 and ds9 • AICR on ds1 and ds9
Experiments • csd site graph • Singular site graph Amod AICR
Experiments Number of clusters
Experiments • AICR • AMod
Conclusions • A novel graph-clustering algorithm is efficient in the exploration of densely interconnected clusters. • A refine clusters may be more densely interconnect than conventional ones.
Comments • Advantages • It's efficient for densely interconnected datasets. • Applications • Hierarchical agglomerative Clustering