This could also be used another approach to recommendation engine. The typical recommendation engine predicts that users who buy butter also buy eggs, but doesn

This could also be used another approach to recommendation engine. The typical recommendation engine predicts that users who buy butter also buy eggs, but doesn't make direct connections between the individual users. The process described in this article instead identifies individual users who act alike, and that can also be used to predict if user A buys an ostrich egg for no logical reason, then their "collusion" peers are also highly likely to buy an ostrich egg ... you can assume if they're not colluding directly then they at least think alike if they have a very high collusion ratio.