RIVERSIDE, Calif. (www.ucr.edu) -- Cyber-crime is expanding to the fertile grounds of social networks and University of California, Riverside engineers are fighting it.
A recent four-month experiment conducted by several UC Riverside engineering professors and graduate students found that the application they created to detect spam and malware posts on Facebook users' walls was highly accurate, fast and efficient.
The researchers also introduced the new term "socware" to describe a combination of "social malware," encompassing all criminal and parasitic behavior on online social networks.
Their free application, MyPageKeeper, successfully flagged 97 percent of socware during the experiment. In addition, it was only incorrect - flagging posts of socware that did not fit into those categories - 0.005 percent of the time.
The researchers also found that it took an average of .0046 seconds to classify a post, which is far quicker than the 1.9 seconds it takes using the traditional approach of web site crawling. MyPageKeeper's more efficient classification also translates to lower costs, cutting expenses by up to 40 times.
"This is really the perfect recipe for socware detection to be viable at scale: high accuracy, fast, and cheap," said Harsha V. Madhyastha, an assistant professor of computer science and engineering at UC Riverside's Bourns College of Engineering.
Madhyastha conducted the research with Michalis Faloutsos, a professor of computer science and engineering, and Md Sazzadur Rahman and Ting-Kai Huang, both Ph.D. students. Rahman presented the paper outlining the findings at the recent USENIX Security Symposium 2012.
During the four-month experiment, which was conducted from June to October 2011, the researchers analyzed more than 40 million posts from 12,000 people who installed MyPageKeeper. They found that 49 percent of users were exposed to at least one socware post during the four months.
"This is really an arms race with hackers," said Faloutsos, who has studied web security for more than 15 years. "In many ways, Facebook has replaced e-mail and web sites. Hackers are following that same path and we need new applications like MyPageKeeper to stop them."
The application, which is already attracting commercial interest, works by continuously scanning the walls and news feeds of subscribed users, identifying socware posts and alerting the users. In the future, the researchers are considering allowing MyPageKeeper to remove malicious posts automatically.
The key novelty of the application is that it factors in the "social context" of the post. Social context includes the words in the post and the number of "likes" and comments it received.
For example, the researchers determined that the presence of words - such as 'FREE,' 'Hurry,' 'Deal' and 'Shocked' - provide a strong indication of the post being spam. They found that the use of six of the top 100 keywords is sufficient to detect socware.
The researchers point out that users are unlikely to 'like' or comment on socware posts because they add little value. Hence, fewer likes or comments are also an indicator of socware.
Furthermore, MyPageKeeper checks URLs against domain lists that have been identified as being responsible for spam, phishing or malware. Any URL that matches is classified as socware.
During the four-month experiment, the researchers also found:
A consistently large number of socware notifications are sent every day, with noticeable spikes on a few days. For example, 4,056 notifications were sent on July 11, 2011, which corresponded to a scam that went viral conning users into completing surveys with the pretext of fake free products.
Only 54 percent of socware links have been shortened by URL shorteners such as bit.ly and tinyurl.com. The researchers thought this number would be higher because URL shorteners allow the web site address to be hidden. They also found that many scams use somewhat obviously "fake" domain names, such as http://iphonefree5.com and http://nfljerseyfree.com, but users seem to fall for it and click the link.
Certain words are much more likely to be found in Facebook socware than in e-mail spam. For example, "omg" is 332 times more likely to appear in Facebook socware. Meanwhile, "bank" is 56 times more likely to appear in e-mail spam.
Twenty percent of socware links are hosted inside of Facebook.
This activity is so high that the researchers expect that Facebook will have to do more to protect its users against socware.
"Malware on Facebook seems to be hosted and enabled by Facebook itself," Faloutsos said. "It's a classic parasitic kind of behavior. It is fascinating and sad at the same time."