A structural approach to community-level social influence analysis

Social communities shape the way people interact. E.g. members of online discussion communities frequently exchange information, experience, or knowledge about practically anything from software to bird watching. The rising availability of data from social communities has led to a surging research interest in their modelling and analysis. One of the main motivations behind the interest is the promise that models of community dynamics may help the stakeholders to make good use of the time or capital they invest to the communities. A prominent problem in the study of communities has been to quantify and explain how their members influence each other. This has found many applications in various areas such as public health promotion programs or viral marketing. However, how a community, as entity, influence or is influenced by other communities, i.e. cross-community influence, has been less studied. We propose that the relationships a community, as a whole, maintains with other communities contribute to how it evolves in terms of growth, topic, and decline. The main problem we address is the measurement, analysis, and explanation of influence relationships between various types of social communities. We address the problem by developing a computational model for cross-community influence that we call COIN. Our model is flexible and caters for differences between data from various types of communities. The core of COIN is based on a purely network-based representation of social interactions. COIN can thus reveal and explain influence relations between communities for which no additional data like textual content is available due to e.g. legal reasons. However, we also devise an extended version that integrates and helps to interpret additional information about the interactions extracted from textual data. The model is evaluated on three data-sets from leisure, business, and scientific communities. We present and explain a broad range of cross-community influence phenomena. We describe a rise of global authorities or communities that act as a hub, as well as dynamic patterns of influence between pairs of communities. Furthermore, we demonstrate how can be the cross-community influence exploited for efficient information diffusion. Last but not least, we use COIN to identify scientific communities that became increasingly isolated, self-referential, and shrinking, thus shedding more light onto the possible causes of a community's decline.

Funder

Science Foundation Ireland (SFI)
European Commission

Rights

Attribution-NonCommercial-NoDerivs 3.0 Ireland

cbn

Collections

University of Galway Theses (PhD Theses)