A distributed approach to cluster multi-view relational data

Abstract

Clustering of multi-view data has become an important research field. The efficient clustering of multi-view data is a challenging problem. This work aimed to investigate a distributed approach to cluster multi-view relational data. A PSO-based hybrid method was used to generate clustering from all views independently. Five different objective functions were explored to induce diversity to the clusterings since each function looks for different cluster structures. Five different consensus functions were compared to produce the final partition from the ensembles. Three multi-view real-world data sets were considered in this study. The Adjusted Rand Index, the F-measure and Silhouette clustering validity indexes were used to assess obtained clusterings. The distributed approach found better clusterings for all data sets considering at least one consensus function.