Views: 0 | Downloads: 0
Network analysis is becoming increasingly important for understanding and analyzing networks
of social interactions. This thesis unifies work on developing methods for analyzing
networks using structural, textual and temporal data modality.
The first topic of our research is development of an algorithmic approach for identifying
important subsets of nodes in the sense of cardinality of the union of neighborhoods. We
describe the polynomial algorithm for finding the group with maximal degree centralization,
with the best-possible approximation ratio. The constructed algorithm is tested on six
real-world networks. In experiments we observe a property of unimodality of group degree
centralization for group size parameter.
We also propose an approach to using the network structure as captured by the coreperiphery
partition, to study evolution of social networks over time. We use machine learning
techniques on one side and analyze parameters of ego-networks to reveal the differences
between individuals in the cores on the other. In particular, we analyzed membership in
the cores of the scientific coauthoring networks, dynamics of individuals related to the
network cores, and ego-networks of all individuals.
Another important aspect of our research is investigating interdisciplinarity of scientific
fields based on graph of collaboration between the researchers. A new measure for network
diversity is proposed that takes into account graph content and structure.
Finally, we propose an approach for conceptualizing science using collaboration and
competences of the scientists. The approach is demonstrated on Slovenian national research
data, in a publicly available web application named Science Atlas.