User study on illicit casino promotion

We conducted a user study on casino promotion and got similar results to pharmacy promotion. It further demonstrated the strong generalization capability of our attack method.

We used a similar sampling rate to the user study in Section 4.4 and randomly selected 10 malicious articles generated by MAWSEO and 10 Wiki benign articles from WikiDump. We also included 15 articles generated by HotFlip, Collision, and PAT. In total, our article pool contains 35 unique articles. 

We received 39 valid response, including 54 valid answers to malicious articles, 59 valid answers to baseline articles, and 43 valid answers to benign articles. 81.5% of malicious articles were considered trustworthy, while 90.7% in benign articles and 16.9% in baseline articles. The survey result indicates that the articles modified by MAWSEO are highly trustworthy and have the potential ability to bypass Wiki users' awareness of malicious modification.

Participants reached promotional content in 77.8% of malicious articles, compared to 25.4% of baseline articles. It shows that participants can receive and accept promotional content in the malicious articles better than in the baseline articles.

In the reviewer acceptance study, 53.8% of revisions on malicious articles were accepted, while 0% on the baseline. It indicates that the malicious articles have the potential capability to pass the Wiki reviewers' vandalism checking.