Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Computer Science


Dr. Lu Xiao

2nd Supervisor

Dr. Robert Mercer

Joint Supervisor


Given the ever-growing information generated through various online social outlets, analytical research on social media has intensified in the past few years from all walks of life. In particular, works on social Web intelligence foster and benefit from the wisdom of the crowds and attempt to derive actionable information from such data. In the form of collective intelligence, crowds gather together and contribute to solving problems that may be difficult or impossible to solve by individuals and single computers. In addition, the consumer insight revealed from social footprints can be leveraged to build powerful business intelligence tools, enabling efficient and effective decision-making processes. This dissertation is broadly concerned with the intelligence that can emerge from the social Web platforms. In particular, the two phenomena of social privacy and online persuasion are identified as the two pillars of the social Web intelligence, studying which is essential in the promotion and advancement of both collective and business intelligence.

The first part of the dissertation is focused on the phenomenon of social privacy. This work is mainly motivated by the privacy dichotomy problem. Users often face difficulties specifying privacy policies that are consistent with their actual privacy concerns and attitudes. As such, before making use of social data, it is imperative to employ multiple safeguards beyond the current privacy settings of users. As a possible solution, we utilize user social footprints to detect their privacy preferences automatically. An unsupervised collaborative filtering approach is proposed to characterize the attributes of publicly available accounts that are intended to be private. Unlike the majority of earlier studies, a variety of social data types is taken into account, including the social context, the published content, as well as the profile attributes of users. Our approach can provide support in making an informed decision whether to exploit one's publicly available data to draw intelligence.

With the aim of gaining insight into the strategies behind online persuasion, the second part of the dissertation studies written comments in online deliberations. Specifically, we explore different dimensions of the language, the temporal aspects of the communication, as well as the attributes of the participating users to understand what makes people change their beliefs. In addition, we investigate the factors that are perceived to be the reasons behind persuasion by the users. We link our findings to traditional persuasion research, hoping to uncover when and how they apply to online persuasion. A set of rhetorical relations is known to be of importance in persuasive discourse. We further study the automatic identification and disambiguation of such rhetorical relations, aiming to take a step closer towards automatic analysis of online persuasion. Finally, a small proof of concept tool is presented, showing the value of our persuasion and rhetoric studies.