The research on opinion dynamics in social networks and opinion influence models often suffer from a lack of grounding in social theories as well as deficient empirical data validation. The current availability of large datasets, and the ease we can now collect social data from the Internet, makes validation of theoretical social models a less difficult task. Starting by a state-of-the-art of the research and practice concerning political opinion dynamics in social networks, we identify the main strengths and weaknesses of this domain. We then propose a novel method for uncovering political opinion dynamics using on-line data gathering. The method includes three distinct phases: (1) data collection, (2) multi-agent modelling (3) validation. Specifically, we tested the significance of both Social Impact Theory, originally proposed by Latané (1981), and Brownian Agent modelling, proposed by Schweitzer (2002), for characterizing political opinion formation during electoral periods. These two models were tested using more than 100.000 tweets collected during the periods from the 30th of October to the 21st of January 2011 and from the 27th of March to the 6th of June 2011, concerning the Portuguese presidential and legislative elections occurred in 2011. Following the data collection, two distinct on-line communities were inspected: the general Twitter user community, and the traditional news media Twitter feeds. The opinion dynamics was simulated with grid adjustment of model parameters. This operation was performed on separate empirical series, respecting the talk about the six electoral candidates and parties. The complete process allowed concluding about the explanatory power of Social Impact Theory and Brownian Agents, and, on the other side, allowed characterizing opinion dynamics in this specific case study. This article details each phase of the method, illustrated using the dataset available at http://work.theobservatorium.eu/presid2011.

--

Opinion dynamics,Web 2.0,Social networks,Multi-agent simulation,Sociophysics,Econophysics