Finding essential connections in a network of time-varying interactions

Nowadays, time-resolved data about human behavior and interactions is becoming increasingly accessible. In particular, many data take the form of networks evolving over time, such as networks of interactions or communications between individuals. These data yield important progresses in the understanding of social phenomena as well as in the study of data-driven models describing the propagation of information or infectious diseases during contacts between people.

In these complex data, it is however often difficult to separate the most important and crucial links from those occurring simply by chance. For example, in a network of interactions between people, the number of interactions between two individuals depends on their social activity: even if people interact at random, two very active individuals would have more interactions than two individuals less socially active. Thus, simply counting the number of interactions is not enough to find the most relevant links and filter out non-essential links.

In this context, ISI Senior Researcher Alain Barrat ( Center for Theoretical Physics, CNRS & Aix-Marseille Univ) and two colleagues from Kobe University and LINE company in Japan have developed a new network filtering method to identify the essential connections of a temporally evolving network, which they call “significant ties”. A paper just out in Nature Communications shows how this method controls for the differences in the levels of social activity of individuals and, for the first time, takes into account the temporal evolution of the network: the methods published so far had indeed been developed for static networks.

The team applied this method to data describing different systems of socio-economic relevance (people-to-people contacts, inter-bank loan networks, email exchange networks ...). Researchers have found that in populations structured in groups (such as interaction networks in schools), the significant ties are predominantly within the groups, whereas the links between groups are compatible with random meetings.
An important advantage of the new method lies in its ability to find not only important interactions but also more complex significant structures, such as triads of individuals interacting simultaneously more than expected given their social activity. The above methods, which did not take into account temporality, could only define such structures as superpositions of significant ties; interestingly, the new published results show that many significant triads are not made of three significant dyads, and hence would not be found by previous methods.

Future applications of this method include more detailed studies of time-varying data, the definition of new representations of complex data, the use of these representations in data-driven models of processes on networks, the comparison of data describing different contexts, as well as potential uses in studies of behavior within social groups.

The Structured Backbone of Temporal Social Ties”, Teruyoshi Kobayashi, Taro Takaguchi, Alain Barrat, Nature Communications, 15th January 2019