Trust and Credibility in Online Social Networks

Abstract

Increasing portions of people's social and communicative activities now take place in the digital world. The growth and popularity of online social networks (OSNs) have tremendously facilitated online interaction and information exchange. As OSNs enable people to communicate more effectively, a large volume of user-generated content (UGC) is produced daily. As UGC contains valuable information, more people now turn to OSNs for news, opinions, and social networking. Besides users, companies and business owners also benefit from UGC as they utilize OSNs as the platforms for communicating with customers and marketing activities. Hence, UGC has a powerful impact on users' opinions and decisions. However, the openness of OSNs also brings concerns about trust and credibility online. The freedom and ease of publishing information online could lead to UGC with problematic quality. It has been observed that professional spammers are hired to insert deceptive content and promote harmful information in OSNs. It is known as the spamming problem, which jeopardizes the ecosystems of OSNs. The severity of the spamming problem has attracted the attention of researchers and many detection approaches have been proposed. However, most existing approaches are based on behavioral patterns. As spammers evolve to evade being detected by faking normal behaviors, these detection approaches may fail. In this dissertation, we present our work of detecting spammers by extracting behavioral patterns that are difficult to be manipulated in OSNs. We focus on two scenarios, review spamming and social bots. We first identify that the rating deviations and opinion deviations are invariant patterns in review spamming activities since the goal of review spamming is to insert deceptive reviews. We utilize the two kinds of deviations as clues for trust propagation and propose our detection mechanisms. For social bots detection, we identify the behavioral patterns among users in a neighborhood is difficult to be manipulated for a social bot and propose a neighborhood-based detection scheme. Our work shows that the trustworthiness of a user can be reflected in social relations and opinions expressed in the review content. Besides, our proposed features extracted from the neighborhood are useful for social bot detection.