Invited talk at SSC2016 by Prof. Frank Schweitzer, Chair of Systems Design, ETH Zurich
Computational social science is only as good as the models used and the data analysed to describe and to understand social phenomena.
So, when is a model good enough? And how can the available data be used to calibrate a model? In this talk, I will illustrate these problems by focusing on emotional influence. Different from opinions, emotions are short-lived psychological states that strongly bias individual behavior. Following Russell (1980), emotions can be classified along the dimensions of valence (the pleasure associated with emotions) and arousal (the degree of activity induced by emotions). We can quantify the emotions of individuals who, for example, participate in an online chat, by surveying their subjective response or by providing a sentiment analysis of the text they read and write. But how can this be linked to a model?
We have developed an agent-based modeling framework where the dynamics of individual valence and arousal and the communication between agents is explicitely modeled. For the emotional response of agents, we can test different assumptions (a) by fitting these to the observed subjective response, and (b) by comparing the model output to the observed collective behavior. We will provide different examples (communication in online chats, product reviews, emotional cascades) to demonstrate that the agent-based models can remarkably well reproduce the real behavior of users in online social media.