Analysis and detection of low quality information in social networks

Abstract

Low quality information such as spam and rumors is a nuisance to people and hinders them from consuming information that is pertinent to them or that they are looking for. As social networks like Facebook, Twitter and Google+ have become important communication platforms in people's daily lives, malicious users make them as major targets to pollute with low quality information, which we also call as Denial of Information (DoI) attacks. How to analyze and detect low quality information in social networks for preventing DoI attacks is the major research problem I will address in this dissertation.

Although individual social networks are capable of filtering a significant amount of low quality information they receive, they usually require large amounts of resources (e.g, personnel) and incur a delay before detecting new types of low quality information. Also the evolution of various low quality information posts lots of challenges to defensive techniques. My work contains three major parts: 1). analytics and detection framework of low quality information, 2). evolutionary study of low quality information, and 3). detection approaches of low quality information. In part I, I proposed social spam analytics and detection framework SPADE across multiple social networks showing the efficiency and flexibility of cross-domain classification and associative classification. In part II, I performed a large-scale evolutionary study on web page spam and email spam over a long period of time. In part III, I designed three detection approaches used in detecting low quality information in social networks: click traffic analysis of short URL spam, behavior analysis of URL spam and information diffusion analysis of rumors in social networks. Our study shows promising results in analyzing and detecting low quality information in social networks.