Statistical inference in bipartite networks applied to social dilemmas and human microbial systems

Identifier:  TDX:3359

Handle: https://hdl.handle.net/20.500.11797/TDX3359

Authors:  Cobo López, Sergio

Abstract:
Link prediction in complex networks is a very important problem due to its practical importance. However, the ability of predicting links successfully arises naturally from a good understanding of the functioning and the dynamics of the system under study. In this thesis, we explore the problem of interpretable link prediction in complex networks. In particular, we focus on multilink bipartite networks; first, because bipartite networks are ubiquitous in many natural and social systems and second, because the existence of multiple links allows us to analyze different types of interactions. To that end, we present a family of models that can make interpretable link prediction in this kind of networks and we apply them to two different problems. In the first problem, we consider a social experiment in which a large group of people make strategic decisions in a game theoretical context. We observe that it is possible to find groups of people according to their collective strategic behaviors (i.e., how do they make decisions) and that it is possible to make link prediction upon those groups. In our case we can successfully predict around 75% of the decisions. The second problem is a human microbiology one. We have data on gut microbiome samples from a large number of patients. In a similar fashion, we look for groups of patients according to similarities in their microbial profiles. We then make predictions of microbial abundances using that group structure with an approximately 80% accuracy rate. In conclusion, we show that it is possible to implement our methods to problems that are very different in their nature, so that we can build predictive and interpretable models that work on the ability to identify groups or communities of nodes and track the interactions among those communities.