Clustering of social networks : a managerial tool : comparison of methods

Details

Supervisors

Saerens, Marco

Faculty

Louvain School of Management

Degree label

Master [120] en Ingénieur de gestion

Abstract

The point of this thesis is to implement and test a new clustering technique (the possibilistic fuzzy c-means, PFCM), and to illustrate how clustering techniques can be useful in the business life. This thesis is structured in three parts, to answer the following two research questions: • Does the possibilistic fuzzy c-means algorithm give better results than the distance-based k-means, the kernel k-means and the Louvain method? • Does using a Gaussian kernel-based possibilistic fuzzy c-means yields better results than the “regular” possibilistic fuzzy c-means? The first part of the thesis, “Theoretical background”, covers the theoretical foundations of clustering techniques. It will describe what clustering (and its related notions, such as proximity, similarity and kernels) really is. This section also presents the algorithms that will be tested in the thesis (distance-based k-means, kernel k-means, Louvain method and possibilistic fuzzy c-means clustering). Finally, this sections also details the performance measures used to assess the quality of a partition in clusters. This part also comprises a chapter covering the links between clustering and management. This specific chapter highlights popular or promising applications of clustering in marketing, sales, finance, logistics and business processes. The second part presents the experiments and the results. It starts by defining the procedure that will be applied to perform the clustering tasks and describe the datasets on which the algorithms will be tested. The third part, “further work and conclusions”, gives some ideas for improving the results obtained with this thesis.