Socio-semantic networks as mutualistic networks?
Jonathan St-Onge, Johanne Saint-Charles, Pierre MongeauIn this work, we propose an original way for thinking about socio-semantic networks accompanied by a hybrid methodology. Inspired by recent work done in ecology on mutualist network, we suggest that the link between semantic and social relations can be approached as is the case in ecological networks: agents visit hidden topics in a similar way that insects visit specific plants for pollination. This type of coevolution relationship is mutually defined insofar as one can hardly survive without the other. We use Bayesian methods to estimate if the Enron socio-semantic email network exhibits moderate connectivity, long-tail degree distributions and nestedness, all essential properties that characterize mutual networks. To do so, we build a plant-pollinator matrix where our “insect species” are communities detected via a nested stochastic block model, our “plant species” are hidden topics in text as detected by structural topic model, a type of Latent Dirichlet Allocation, and the interaction between the two are estimated based on the number of visits of a community member to a specific topic. In doing so, we propose a way of thinking and modeling socio-semantic networks as an interaction between two intermingled data generating processes, i.e. a social community generating process and a document generating process. As is the case in situations where we compare cultural with biological dynamics, we also discuss the limitations of our analogy.