On Measuring Social Dynamics of Online Social Media

Abstract

Due to the complex nature of human behaviour and to our inability to directly measure thoughts and feelings, social psychology has long struggled for empirical grounding for its theories and models. Traditional techniques involving groups of people in controlled environments are limited to small numbers and may not be a good analogue for real social interactions in natural settings due to their controlled and artificial nature. Their application as a foundation for simulation of social processes suffers similarly. The proliferation of online social media offers new opportunities to observe social phenomena “in the wild” that have only just begun to be realised. To date, analysis of social media data has been largely focussed on specific, commercially relevant goals (such as sentiment analysis) that are of limited use to social psychology, and the dynamics critical to an understanding of social processes is rarely addressed or even present in collected data. This thesis addresses such shortfalls by: (i) presenting a novel data collection strategy and system for rich dynamic data from communities operating on Twitter; (ii) a data set encompassing longitudinal dynamic information over two and a half years from the online pro-ana (pro-anorexia) movement; and (iii) two approaches to identifying active social psychological processes in collections of online text and network metadata: an approach linking traditional psychometric studies with topic models and an algorithm combining community detection in user networks with topic models of the social media text they generate, enabling identification of community specific topic usage.