How to cluster users in recommender systems to detect an attack?

Question

Clustering

Good afternoon! I am interested in the question of how to identify suspicious accounts using user clustering so that their ratings (underestimated or overestimated) are not counted in recommender systems.
There is enough information on how to divide users according to their interests, but I can’t find how to understand these are real people or this is an attack (for example, someone wanted to underestimate their competitor’s product, but increase theirs).
Share your thoughts or articles.
If the question is not clear, I can rephrase it.
Thanks

Answer 1

the standard deviation of the
upper and lower quartiles
, etc.
conventional statistical methods, well, just cropping according to unrealistic parameters, like a height of 2.5 meters, a comment a minute after the publication of an hour-long video, and so on

Answer 2

The easiest way is to find the average ratio of the number of likes to the number of views for each unique.
And make a width tolerance in the median "corridor" of 50% of all unique users.
Everyone who finds himself outside this "corridor" will be cheaters.

Answer 3

Share your thoughts or articles.
If the question is not clear, I can rephrase it.
Yes, it seems like the intelligent ones have gathered here, so it’s possible not to reformulate. But if you want, then of course you can. In the meantime, you will reformulate the answer briefly, what I know.
I personally do not deal with this task, but I know people who deal with it professionally, i.e. for serious customers. So no one will openly tell you their results - as soon as such information becomes open, there will instantly be especially mentally gifted people who will try to bypass this protection. Who needs it?
And so, the analysis is carried out by conventional methods from the field of Fraud Detection. There are many such methods and tools, books are written on this subject. But this is all a "gentle introduction" to the topic, distant approaches to real cases. Well, as in banking systems, everyone has heard about the methods of catching fraudulent transactions that they write about. It seems like here it is, information available to everyone - deviations, search for anomalies, 3 sigma, spatial gaps, etc. - but how it really works in real banks - alas, "know how" and a mystery with seven seals.