Google-sponsored research struggles with fake reviews

Researchers at the University of Illinois at Chicago have tackled the issue of fake reviews, which they said could seriously damage online business. Google-supported research is aimed at finding organized groups of scam commentators, automating the process of identifying and closing them.



Fake reviewers can have a truly destructive effect on various Internet businesses, but the problem of fake user comments has gained particular importance in socially-oriented services such as Yelp and TripAdvisor, where positive (to support business) and negative (to harm competitors) comments have become real an epidemic.



Researchers say that eliminating fake comments is rather expensive for affected businesses - the process itself is laborious and cannot be automated.

')

In their article , university researchers and a Google employee present the GSRank algorithm, which will help fight counterfeiting user reviews.



The key to identifying groups of organized commentators is their behavior, the article says, the key parameters are:

• Time window - members of a group working on increasing or decreasing a product or service often leave comments in a rather narrow time interval.
• Deviation is different from the natural, as the group is hired to move the rating in one direction or another, and all its posts carry a certain rating. The degree to which the group's estimates differ from the natural background makes it possible to understand that someone is trying to deceive the system.
• Content similarity - the group often uses not only the same ratings, but also the same content for the description. In addition, people who make money on this often have a certain set of phrases that they use over and over again in different reviews.
• In the first place , the researchers also noticed that fake reviews are usually published at the beginning of the product or service life cycle. “Spammers tend to show up before anyone else to get the most impact,” they write. “When employees of a fraudulent group are among the very first commentators, they can form a mood.” Knowledge of such a pattern can help in their identification.
• Group size - the size of the group and its relation to the number of real comments can show the presence of spammers, and it is also difficult to assume that, say, 10 characters will comment on a variety of products over and over again. The presence of an established group, seen in reviews of different products, indicates the presence of malicious intent.

The researchers noted that they can not distinguish between several people working together or one spammer using multiple user logins. However, since the algorithm relies on behavior and not on individuals, this no longer has any meaning.