Uma análise empírica de interações em redes sociais

Abstract

Online social networks (OSNs) have become extremely popular websites and part of our daily lives. OSNs represent a new kind of information network that allow users to interact online. Every day, huge amounts of content are shared and millions of users converse through online social links. Understanding what activities users do, what content they share, and how they interact when they connect to these sites create not only opportunities for better interface and system design, but is also important for many applications, such as advertising, political campaigning, and detection of opportunistic and malicious behavior. Despite all these potential benefits, little is known in the research community about it.In this thesis we provide an in-depth study of user interactions in OSNs, covering aspects of user behavior and navigation across social features as well as aspects of unsolicited content exchanged on social interactions. To do that, we gathered data from actual OSN sites, including YouTube and Orkut. We then study how users interact across a number of OSN features, providing a global picture of the range, duration, frequency, and sequence of activities that users do when they connect to these sites. Second, we provide a characterization of the interactions that emerge from YouTube's video responses, a feature that allows users to interact primarily using videos rather than text. Our study unveils typical user behavioral patterns and identifies novel forms of unsolicited content in OSNs. Some users post video response spam (i.e. video responses unrelated to the topic discussed) in order to increase the popularity of a video, or spread advertisements and pornography. We propose a machine learning-based method that is able to accurately identify the majority of these users.