The work presented in this thesis has been performed within the collaboration among MHPC and the Analytic Solution Center (ASC), an office which is part of Assicurazioni Generali Head Office and is responsible of advanced analytic and big data projects. Generali proposed the overall topic: modelling of the customer database as a social network and community identification.
The Analytic Solution Center took also care to identify such a large dataset within the company and its subsidaries. They finally identify the Genertel (a subsidiary company of Generali) customer-base archive as the dataset to work on. The data have been first anonymized by Genertel and then the anonymized dataset has been delivered for the analysis. The goal of the project is to implement on HPC resources a software able to detect the communities from the dataset identified.
The work performed will be presented in this document as follows. In the rest of this chapter we will give a short introduction to Cluster Analysis and the most important related algorithms. In chapter 2 we will describe the hardware and software architecture we implemented, and we will give more details about the two clustering algorithms we adopted in this work. In chapter 3 we will describe the implementation on HPC platforms of our software architecture, and the whole “optimization journey” we performed. Finally, in chapter 4 we will analyse the performance of our implementations, and the results obtained in terms of network communities identified.
The whole activity was co-supervised by Generali Big data group (dr. Valerio Consorti), which suggested for the greatest part the way to proceed. We remark that some deliverables has been submitted to Genertel and intermediate results were periodically discussed during several meetings.