Accession Number:



Advances in Social Circles Detection

Descriptive Note:

Technical Report

Corporate Author:

Universitat Politecnica De Valencia VALENCIA Spain

Personal Author(s):

Report Date:


Pagination or Media Count:



Social circles arised out of a need to organize the contacts in personal networks, within the current social networking services. The automatic detection of these social circles still remains an understudied problem, and is currently attracting a growing interest in the research community. This task is related to the classical problem of community detection in networks, albeit it presents some peculiarities, like overlap and hierarchical inclusion of circles. The usual community detection techniques cease to be the most appropriate, due to these characteristics. Prediction is performed from two data sources the network graph and node attributes corresponding to users profile features. In this thesis, new approaches to this task are discussed and the results obtained from a thorough experimentation are presented. We provide a review of the state of the art in the fields of community detection in graphs, community detection in social networks and social circles detection. We describe the datasets employed in our experiments, both retrieved from Facebook, and we design a variety of feature representations, both from the structural network information and the users profile information. We define and comment the prediction techniques in which our work is based multi-assignment clustering, restricted Boltzmann machines and k-means. We describe some evaluation measures that have been proposed for social circles detection, and provide a critical commentary of some of them, as they present some flaws which lead to degenerate optimal performance.

Subject Categories:

Distribution Statement: