Consumers' purchase decisions are influenced by user generated online reviews. The potential for posting fictitious reviews that sound authentic is causing concern. Deceptive opinion spam is a growing problem for the online review site communities.
Changing online sentiment is hard enough. But fraudulently working together to try and control the sentiment about a product is just plain wrong.
A new software algorithm was announced last week at the World Wide Web 2012 conference in Lyon, France. It tries to detect groups of spammers working together to influence products.
Opinion spamming is quite common. Have a look at any successful web sites with multiple reviews on the site. Some reviewers try to game the system by promoting or demoting target products. Groups of reviewers can work collaboratively to write fake reviews and can often take total control of the sentiment of the site.
These fake reviewers or spammer groups are hard to detect using review content features or methods to detect abnormal behaviours or patterns. One reviewer could log on with multiple IDs, or there could be multiple reviewers that are paid to write reviews.
Timing is important when spotting spammers. Often spammers post similar reviews within a short time of each other (The Group Time Window) and use very similar language (Group Content Similarity).
But it is hard to detect fraudulent reviews. Have a look at these reviews of the Chicago Hilton Hotel and try to guess which is fraudulent:
1. "My husband and I stayed in the Hilton Chicago and had a very nice stay! The rooms were large and comfortable. The view of Lake Michigan from our room was gorgeous. Room service was really good and quick, eating in the room looking at that view, awesome! The pool was really nice but we didnt get a chance to use it. Great location for all of the downtown Chicago attractions such as theaters and museums. Very friendly staff and knowledgable, you cant go wrong staying here."
2. "We loved the hotel. When I see other posts about it being shabby I can't for the life of me figure out what they are talking about. Rooms were large with TWO bathrooms, lobby was fabulous, pool was large with two hot tubs and huge gym, staff was courteous. For us, the location was great across the street from Grant Park with a great view of Buckingham Fountain and close to all the museums and theatres. I'm sure others would rather be north of the river closer to the Magnificent Mile but we enjoyed the quieter and more scenic location. Got it for $105 on Hotwire. What a bargain for such a nice hotel."
The answer can be found on page 202 of the Online Review Communities document. It is hard to detect between fake and authentic reviews isn't it? It is almost impossible to recognise the spam by simply reading every review. More information is needed to make a good judgement.
Arjun Mukherjee and Bing Liu from the University of Illinois at Chicago collaborated with Natalie Glance from Google which gave a Google Faculty Award to partially support the research.
Spotting Fake Reviewer Groups in Consumer Reviews focuses on the algorithm, GSRank which can consider relationships amongst groups, individual reviewers and products reviewed.
The software algorithm can detect review threads that are trying to get control of sentiment and label the 'spamicity' of the group in order that the group can be ranked and dealt with accordingly.
GSRank significantly outperformed the other algorithms in use, showing that spam reviewers can and will be caught.
Until then, careful moderation of posts to stop bursts of sentiment spamming might be our only option...