Focus on... Kevin Polisano and communities on Twitter

on the June 20, 2019

Kevin Polisano is a postdoctoral researcher in Grenoble Informatics Laboratory (LIG). He works on the link between the diffusion of information and the structure of communities on Twitter, in the framework of the Data Institute’s work package 4 on “Data Science, Social Media and Social Sciences”.

After a PhD in Image Processing at Jean Kuntzmann Laboratory (LJK - UGA) about “Modeling anisotropic texture by monogenic wavelet transform”, Kevin Polisano headed for the data science for his post doc. He wanted to bring his skills to an active field that raises multidisciplinary questions. His postdoctoral research is based on a collaboration with Jean-Marc Francony (Pacte), Eric Gaussier (LIG) and Adeline Leclercq-Samson (LJK).

Kevin Polisano works on Jean-Marc Francony’s database created from tweets and retweets about surrogacy and medically assisted procreation. Then he intervened on the mathematical part by developing a co-evolutionary and multi-scale model. The co-evolutionary dimension means the combination between the dynamic of information diffusion in networks and the dynamic of the links creation in the network. The topology of the network determines the diffusion of the information and the diffusion changes the topology in return.

In a sociological perspective between constructivism and holism, the model will have to adopt at the end of the study a multi-scale approach where the individuals influence the structure of the community and in return the structure influences the individuals in their action. The research raises the following questions: how does the behavior of individuals determine the structure of a community and its dynamics? Inversely, how does the community influence the diffusion of information and the behavior of individuals?

This model also takes into account phenomena specific to social networks, as preferential attachment: for instance, the more we have followers, the more we are known, the more attractive we are, and the more likely we are to have new followers. Also, information diffusion resulting from cascades of retweets stimulates such attachment. This phenomenon can be modeled using a Hawkes process, that is, a self-exciting linear process in which an event changes the intensity of the process and then increases the probability that a new event occurs. By using random processes, the model takes into account the uncertainty of the time when a (re)tweet is emitted. A substantial effort has been devoted to developing a software for such processes that can scale to millions of tweets and hundreds of thousands of users.

The analysis of tweets
and retweets revealed on this graph a polarized space between pro- and anti-assisted procreation. The software allows one to detect communities through the structure of the graph. In addition, with visualization tools, social scientists can produce hypotheses on the way in which communities are structured.

A research perspective would be to take into account what people say on the networks. A textual analysis would give for each tweet and user a sort of signature and define profiles based on the words used during the interactions. Each tweet and profile would then be represented by a vector in the graph. This additional information would improve community detection and predictions. This postdoctoral research should lead to the publication of an article and could result in applications on social networks.


Published on June 20, 2019