The role of confirmation bias in the emergence of echo chambers: a data-driven approach

The thesis focuses on the modeling of information diffusion and opinion dynamics on online social networks, using a datadriven approach accounting for the presence of social contagion and cognitive effects such as confirmation bias, cognitive dissonance, and backfire effect. We a) analyze data from a sample of 1.2M of users on the Italian Facebook pages, focused of scientific and conspiracy contents, by means of quantitative methods, statistical analysis, and sentiment analysis; b) develop a data-driven percolation model of signed edges to mimic the information spreading and a theoretical model of opinions formation. We provide evidences that the diffusion of information, either substantiated or not, is promoted by confirmation bias and homophily. This process in turn generates and fosters the formation of homogeneous polarized clusters, the echo chambers. Users’ emotional behavior seems to be affected by their engagement within the community. An higher involvement in the echo chamber, resolves in a more negative emotional state. Lastly, we develop a model of opinions formation that takes into account both confirmation bias and social influence as triggering factors for the group polarization on social networks. Our model is able to reproduce the dynamics we observed on Facebook.