Fake opinions are wide-spread on social media. To ensure that the social media information can be trusted, we must detect such fake opinions and other forms of disinformation. In terms of applications, since social media posts are increasingly used by businesses, organizations, social sciences, and health sciences, detecting disinformation or spamming becomes very urgent. In this project, our goal is to propose a collective detection framework via synergistic integration of multiple information sources: linguistics, user behavior, and network effects to detect fake reviews and fake reviewers (also called opinion spammers). We have used this framework in our proposed algorithms. My group at the University of Illinois at Chicago pioneered the research area of detecting fake reviews and reviewers by publishing the first paper on the topic in 2008. Over the years, we have contributed many useful algorithms. This project continued to contribute significantly to this effort. Below, we highlight three major outcomes or discoveries of this project.

1. Bimodal posting rate: For the first time, we discovered that reviewers’ posting rates (number of reviews written in a period of time) follow an interesting distribution pattern, which has not been reported before. That is, their posting rates are bimodal. Furthermore, multiple spammers tend to collectively and actively post reviews to the same set of products within a short time frame, which we call co-bursting. We also found some other interesting patterns in individual reviewers’ temporal dynamics and their co-bursting behaviors with other reviewers. Inspired by these findings, we proposed a two-mode Labeled Hidden Markov Model to model spamming using only individual reviewers’ review posting times. We then extend it to the Coupled Hidden Markov Model to capture both reviewer posting behaviors and co-bursting signals to detect review spammers.
2. Changed-hand account: As many review hosting sites are actively detecting fake reviews and reviewers (spammers), it has become difficult to post fake reviews. Spammers began to buy reputable accounts to post fake opinions to avoid detection, or to raise their own accounts by behaving like genuine reviewers for a period of time and then use these “trustworthy” accounts to launch spam campaigns. Such accounts are called raised accounts (like raising a child) which behave normally in a period of time to gain credibility but are then used later to write fake reviews or opinions. Before this research, this problem was not studied by the research community. This project conducted a comprehensive study of this new type of spammers (or spam accounts) and proposed an effective technique for their detection. We call such accounts changed-hand accounts.
3. Hidden campaign promoter: Some spammers work together to secretly promote some target products or services by influencing people’s behaviors/opinions/decisions in a latent manner so that the readers are not aware that the messages they are seeing are strategic campaign posts aimed at persuading them to buy some target products/services. Readers take such campaign posts (often opinionated) as just other organic posts from the general public. It is thus important to discover such campaigns, their promoter accounts, and how the campaigns are organized and executed. In this project, we studied this problem on the Twitter platform. Given a set of tweets streamed from Twitter based on a set of keywords representing a particular topic, our technique can identify user accounts that are involved in promotion. We formulated the problem as a relational classification problem and solved it using typed Markov Random Fields (T-MRF), which is a generalization of the classic Markov Random Fields.

We believe that these discoveries and associated detection techniques are all very useful in practice and will help ensure the social media to be a trusted source of information.

Last Modified: 12/01/2018
Modified by: Bing Liu