User participation and community formation in peer production systems

This thesis explores the phenomenon of user participation in peer production communities. In large-scale collaboration communities such as wikis, open source software development teams, and file-sharing groups, members are, in general, not remunerated for contributing to the digital common, so the incentives to participate in these projects are usually construed in terms of extrinsic benefits such as building reputations, reciprocity, gratification, and providing a sense of membership to a socially cohesive group. In fact, like any social group, a community of commons-based peer production is endowed with its own norms, beliefs, and cultural features. In these systems such features, however, are themselves the result of a bottom-up process of cultural formation and opinion aggregation, mostly related to various aspects of the social production of digital content. Unfortunately it is still poorly understood how this process happens – considering that most interactions between members are mediated by the digital artifacts (encyclopedic articles, source code, etc.) that the project is producing. To this end, we studied the process of community formation in a large peer production community by means of statistical analysis and agent-based modeling and simulation. In the first part of this thesis we analyzed the activity of registered editors from the communities of five of the largest versions of Wikipedia, the free online encyclopedia. We found that the distribution of the user activity lifespan is distributed according to a mixture of log-normal distributions. In order to understand these empirical patterns we developed, in the second part of the thesis, an agent-based model of a peer production community. Social influence is modeled in terms of dyadic user-page interactions, under the form of the bounded confidence rule. Thus a measurable aspect of user participation – activity lifespan – is linked to microscopic features of the dynamics of social influence. In order to study the behavior of our model, we perform a factor screening via global sensitivity analysis. We then calibrate our agentbased model using the empirical data from Wikipedia. To this end, an indirect