Opleiding Informatica Algorithms for detecting and analyzing multiplex motifs in large - scale corporate networks

Network analysis is a frequently applied method of studying relations in a dataset. Many network measures assume that each edge in the network conveys the same information. This allows us to express measures such as distance, but can also lead to an incomplete or incorrect view of the data as only one relation in the data can be taken into account. In this thesis we study motif detection on multiplex networks, in which multiple types of interaction occur at the same time. Motifs are subgraphs that occur more frequently in an empirical network than they would do in a synthetically created network. As such they are the basic building block of the network, and can express important structures for a specific type of network. We create a multiplex network from a corporate database by joining two uniplex networks: an ownership network and a board interlock network. By comparing the motifs from uniplex networks to multiplex networks we show the difference in information between the two types of networks. To do so we first define what type of motif is needed to best capture important corporate structures. Then we extend that definition to multiplex motifs. We augment an existing algorithm for motif detection to multiplex motif detection and compare the frequency of motifs in the empirical network to their frequency in a collection of synthetic graphs. We find that multiplex motifs provide information on certain industry sectors, as several motifs contain significantly higher concentrations of a certain sector compared to the full dataset.