Machine Learning is now able to detect ureliable Facebook pages

Facebook pages have become an effective tool via which companies and individuals reach out to their target audience for various marketing and advertising needs. The ease with which a page can be created on Facebook means users can use it as they please; after all, it provides an easy avenue for them to advertise their services and products to people with zero charges.

The ease to create pages has led to many malicious pages springing up on a daily basis. There is the need to detect such harmful pages and, in turn, curb malicious activities on Facebook.

The necessity to limit the actions of some of these malicious pages has led researchers to come up with ways via which deceptive pages can be detected on, not just Facebook, but on all social media outlets. At the forefront of this endeavor is a researcher from the Mahasarakham University in Thailand with the name Panda Songram, who has investigated the use of supervised machine learning to detect the authenticity of a Facebook page.

To carry out her experiment, Songram programmed a machine learning tool to analyze essential page features, like page details, various info about the service or product on the page, general user responses, and actions and behavior of the page admin of several Facebook pages. After analyzing all these features, the machine groups the page into two categories: Unreliable and Reliable.

"To begin, we selected pages at random, and we asked 5 users to label them," Songram mentioned. "The Information of these pages was obtained via the Facebook Graph API. After that, the page features were extracted and investigated."

Songram used a different classifier to determine which pages are reliable and unreliable. From the classifier tested, KNN had come out on top with 88.67% accuracy. Songram also analyzed the Facebook page features to grasp what makes a page reliable or unreliable.

"A telling factor for unreliable pages is the fact that there is a huge gap between the date of the last post and retrieved date, and also the weekly number of posts is small," Songram mentioned. "What this tells us is that, whereas reliable pages are active, unreliable pages are not."

Songram also found out there was a relatively small number of persons discussing unreliable pages online when compared to the numbers of those on reliable pages. The reason for this might be due to the fact that most people find out certain pages are unreliable, and as such, online chatter about those pages dies down with time. On the other hand, reliable pages have far more beneficial URLs and more useful information than those of unreliable pages.

From her analysis, she was able to attain 91.37% accuracy using what she termed "Top 10 features to determine Facebook page reliability." It is with the hope that her findings will be worked on and used to develop a more useful tool that can detect unreliable Facebook pages on the go.