Bob Holmes, consultant
Looking for online product reviews before you buy holiday gifts? Watch your step. Many of those "user reviews" may come from professionals paid to post favourable comments about their employers' products and denigrate their competitors', says a computer scientist who worked undercover in the industry.
Fortunately, though, there may be a way to spot the fakery automatically.
Cheng Chen of the University of Victoria, Canada, worked as a paid poster in China's "Internet water army", so-called because its soldiers flood websites with posts about particular products. In a paper posted on arXiv, Chen and his colleagues describe how project managers organise teams of paid posters, supplying them with comments and video clips to post, and setting rules for when and how often to post, so that they avoid appearing part of a coordinated campaign.
To see if they could recognise paid posters despite this deception, Chen's team focused on online comments relating to a dispute between two Chinese antivirus companies. The researchers sifted through two months' worth of comments on one of China's leading internet news sites and pulled out 552 users who commented on the antivirus companies. Based on his experience in the industry and the contents of their posts, Chen suspected 70 of these might be paid posters.
These suspected paid posters had a higher proportion of new comments (as opposed to replies), posted more often but for a shorter period of time, and were more likely to post similar comments several times than posters not suspected of being paid.
Sure enough, when the researchers applied these criteria to comments on a second news site, the suspected paid posters they flagged matched Chen's subjective classification with a false-positive rate of 1 per cent and a false negative rate of 10 per cent. With a little more refinement, the algorithm could lead to software that screens comments automatically, they say.
http://www.newscientist.com/
No comments:
Post a Comment