On the Correlation between Topic and User Behaviour in Online Communities

In the advancement of a better understanding of user behaviour on the Internet, clustering approaches are used as a means to group and categorise online communities. Although clustering approaches indicate similarity between communities (e.g. through an a priori defined distance function based on user behaviour features) they do not show what causes the differences between the types of communities. In this paper, we study the effects that certain types of topics may have on user behaviour. We cluster 29 online communities of the Stack Exchange platform, and we show the correlation between the clustered user behaviour and the three latent topic properties Accessibility, Sociability and Controversy.