c******n
发帖数: 4965 | 1 【 以下文字转载自 SanFrancisco 讨论区 】
发信人: creation (努力自由泳50m/45sec !), 信区: SanFrancisco
标 题: technical question: click fraud detection?
发信站: BBS 未名空间站 (Tue May 28 01:16:46 2013, 美东)
we are working on some ppc click fraud detection project.
a very hard problem is how to find those clicks which come from "get paid to
click on ads" users, or simply from bots on compromised machines.
for bots, the bots program can decorate their http requests any way they
like, they can write fake cookies, fake referrer, fake landing urls. a lot
of the research on this subject focuses on the frequency per cookie, or try
to look at referrers (reverse spidering), or look at the user's browsing
history (landing url) . but if the bot can fake all these, the analysis on
these features also become useless, right?
the only thing that can not be faked by the bots is the ip. so we have the
very basic, very common frequency per ip analysis. but this is not
effective at all if the fake requests are widely distributed on a
compromised botnet of machines.
anybody working in this field could give some pointers? thanks |
|
