Options
Extracting maximally connected sub-graphs used for community generation
Date Issued
01-12-2011
Author(s)
Isaksson, Lennart
Kumar, Peeyush
Saravanan, M.
Abstract
The ability to mine and extract dense sub-graphs from massive mobile call graphs has become one of the most challenging problems in an exponentially growing mobile telecommunication scenario. In order to meet these challenges,we extracted sub-graphs from massive mobile data on a daily basis chose on a suitable time slot to select a stable graph which generalizes the dynamics of the call graph. Usage of data mining algorithms can support for community generation. In this paper, we use spectral clustering algorithm with additional heuristics to segment the maximally connected sub-graph into knitted subunits (communities). Identified communities are well balanced with a good cut score. The results obtained are better than previous state-of-art machine learning algorithms executed on Power Law Graphs (a characteristic for Mobile Social Network). Further we discuss our inference from the results obtained that the possibility of a node being associated with more than one community is an intrinsic property for social networks. Then we extend to mine the characteristics of identified communities. For this purpose, we analyze the data from Voice and Short Message Service (SMS) services of a telecom networks to incite the structural features of the services. A suitable adaptation of proposed algorithms brings out a good deal of useful information to mobile telecommunication (Telephone Company) operators for their social marketing activities. © 2011 IADIS.