Communities in co-membership networks : Twitter applications
Mohamed Benabdelkrim · HAL (Le Centre pour la Communication Scientifique Directe) · 2021
Graphs are mathematical tools for modelling relationships between entities. They are used in multiple scientific fields: in computer science for knowledge representation, in sociology for studying relationships between people, in chemistry for representing molecules and in many other fields. One specific type of relationships, that is of interest in this work, is the one that represents common membership of a set of entities to a set of groups. This relationship defines a multilayer network called the co-membership network where entities are represented by nodes and groups correspond to layers. Such relationship also has the property of transitivity which makes the layers of the co-membership network fully-connected. Moreover, in many cases, textual descriptions are associated to the groups and are represented in the co-membership network by layer attributes. The objective of this thesis if to develop community detection tools for co-membership networks using both the structural and semantic information they contain. The studied common membership relationship has various applications including the analysis of common attendancy of researchers to conferences and the representation of co-authorship of scientific publications. In social networks, this type of relationship is even more frequent and can be found in Twitter with common membership to lists. In this work, we give a formal definition of co-membership networks and underline the properties that distinguish them from the other multilayer networks. We present two algorithms for delimitting subgraphs that correspond to domains of interest of users. We also explain how to adapt data mining methods for the extraction of itemset and numeric patterns to the task of community detection in co-membership networks. Results are compared to those of state the art methods using two example networks from the social network Twitter and the collaborative encyclopedia Wikipedia.