Below is the abstract along with some visualizations (click images) from my master thesis. If this peaks your interestst the thesis is
freely available here
The SensibleDTU dataset holds, for approximately 1000 participants, detailed information on social
media and face-to-face interaction, highly resolved in time. These very different channels of
data, derived from the Facebook API and mobile-phone Bluetooth scans respectively, are used to
build temporal multiplex networks. The goal of this thesis is to investigate the similarities and differences
between the two modes of data through the lens of dynamic community detection.
First, a general overview of complex networks theory is provided, then the focus is narrowed to a
review of the rapidly expanding subfield of community detection. Further zooming in, clustering
in temporal multiplex networks is treated, specifically the information theoretical framework of the
map equation used to analyse the networks.
Different schemes for coupling strength between multiplex layers are investigated, and their effects
on community detection studied. The results are interpreted through both quantitative measures
and visualisations, focusing on the temporal development of: 1) Detected communities, 2) Community
affiliation of participants and 3) The network as seen from the perspective of a single user.
Effective community detection in the Facebook network is found to be difficult, due to the many
different time-scales over which people communicate. The individual networks display a low average
clustering coefficient, leading to unintuitive partitions.
For the face-to-face network, an optimal interlayer coupling is found, allowing for the distinguishing
of different types of meetings revolving, for example, around university courses, lunch and
evening activities.
In order to study recurring meetings, a network of temporal communities linked by similarityweighted
edges is constructed. Performing graph clustering on such a network yields a partition
into sets of communities with high self-similarity. It is shown how plotting timelines of when the
recurring communities are active can give interesting insight into the social contexts which they
represent.
