Welcome to WebmasterWorld Guest from 18.206.194.210

Forum Moderators: buckworks & eWhisper & skibum

Message Too Old, No Replies

Measures to Verify Genuine Click

discussion related to fake clicks

     
12:52 am on Dec 10, 2005 (gmt 0)

New User

10+ Year Member

joined:Jan 19, 2005
posts:9
votes: 0


Hello Every One :

I came across this today.

[thewhir.com...]

It appears that at times there are too many fradulent clicks that eat up our budget but they are all fake.

We need to see the precautions or ways to avoid fake click, Please discuss the measures to verify a genuine click.

AQ Azhar!

1:43 am on Dec 10, 2005 (gmt 0)

Senior Member

WebmasterWorld Senior Member jtara is a WebmasterWorld Top Contributor of All Time 10+ Year Member

joined:Nov 26, 2005
posts:3041
votes: 0


Ah, transparency or no transparency?

Google doesn't have a great record on transparency. That, or they are just really bad at communicating clearly.

On one hand, if they talk about what they are doing to filter out fradulent clicks, they give it away to the bad guys.

On the other hand, if they do, they give advertisers some basis to evaluate the effectiveness of their efforts. And they open things up for more eyes to find weak spots.

The same debate has ranged on in computer security for years. Personally, I'm for transparency.

I'm not sure that the problem is solvable. It might be that an AI approach is the way to go, and may well be what they are doing.

Somehow, get a good-sized sampling of guaranteed genuine user activity. Maybe agree with some big corporations that are really on top of keeping their machines clean from viruses and trojans to capture their user's browsing habits.

Big privacy issues, there, though. I think you have to tell employees that their browsing behaviour is being studied in this way. And then by simply doing so, you run the risk of skewing the results because the users know they are being studied.

And you also invite unscrupulous employees to inject click-fraud traffic, which they might just get past the corporations anti-virus, anti-trojan efforts.

It would be worth a lot of money to a click-fraud cartel to skew the data for such a system from the get-go. (Think there aren't big organized-crime interests in this? Think again.)

So, you really have to keep it secret. But you've got real ethical and possibly legal problems if you do keep it secret.

Catch-22.

How you get a sampling of virus-free, trojan-free home users is anybody's guess. Just finding ones who care enough about security to BE virus-free/trojan free is going to skew the data.

Catch-22 again.

But assuming you can get over those hurdles - now you throw the fraud-free traffic at an AI engine, and let it learn. And then you throw unfiltered traffic at it and let it learn that as well. Now, the AI engine should be able to differentially figure out what is genuine normal activity, and what is click-fraud.

Only thing is, how accurate will it be?

I once worked for a company that is responsible for generating most of our credit scores. Using AI. Now, there's a scary thought...

Anyway, I was on a project to apply the same technology to web form-filling. A simple time-saver for users. (Forget for a moment what a dumb idea this is in the first place, and the fact that there persist to this day several products that purport to do this. Why?)

Should be a piece of cake, right? I mean, they use this stuff to come up with credit scores. Must be good stuff.

Problem is, in form filling, you need to have nearly 100% accuracy to be useful. Otherwise, it is just to annoying to the user. If they can't rely on it to be almost 100% accurate (like 99.9) then they are going to stop and have to check every field for accuracy. So, they might as well have just typed-in the data in the first place.

The question is, what % accuracy is needed to filter out fradulent clicks? Will 90% do it?

I'm being overly-broad with the concepts here, I'm afraid. Form-filling has some pretty obvious clues in the data. Most fields have recognizable english-language variable names and/or labels. Last names tend to come right after first names. A little field in the middle tends to be a middle initial. Pretty straight-forward stuff.

Now, if AI can do form-filling at 90% accuracy, do you think it can filter out, say 75% of fradulent clicks? Or maybe just 20%, given that it's a much more difficult pattern to spot?

I suspect that AI is the solution. I'm just not sure we're going to see it in the very near future.

Of course, the data would be pre-filtered. There are plenty of obvious specific cases that can be filtered out ahead of the AI. Repeated clicks of the same link within a certain time period. A high density of clicks on revenue-producting links. You can filter out clicks just coming out of the blue from random proxy servers or random offshore workers or bots that have been given a simple list of URLs to visit, because they are isolate and so they are obvious.

Even so, what is left is going to be awfully tough to filter with today's technology.

7:13 am on Dec 12, 2005 (gmt 0)

New User

10+ Year Member

joined:Jan 19, 2005
posts:9
votes: 0


Thanks alot jtara! For your time and very informative reply.

But still I'm curious to know how advertiser can avoid fake clicks immediately to save his budget. I want to know with reference to advertiser.

Thanks again!

AQ Azhar!

12:04 pm on Dec 12, 2005 (gmt 0)

Junior Member

10+ Year Member

joined:Jan 6, 2005
posts:106
votes: 0


I find it quite an interesting area, and have assumed it surely can only really be affecting the big players? A company with the money, effort and resources to hire a team/designated people to click and run click fraud software must have some cash to spend attmepting the bring their competitors down. Also, they are assuming that their competitors are stupid enough to waste all their budget on underperforming campaigns (i.e. big players, waste more as they spend more) not much point click frauding the local guitar shop with his weekly adwords limit of 50 for example, and checks his ad performance daily.

The alternative is a long, slow but slight click fraud, knocking off dollars a day, attempting to cut the profits of competitors with a small amount of un-notice-able clicks?

So have always kind of though it doesn't affect, me more the likes of the multi-million pound advertisers. But maybe I'm worng :p

12:20 pm on Dec 12, 2005 (gmt 0)

Senior Member

WebmasterWorld Senior Member 10+ Year Member

joined:Sept 25, 2004
posts:2156
votes: 0


Hi,

I think there's also significant low-level click fraud by publishers trying to boost their income. Maybe they also steal candy from babies when hungry...

I've had to stop using content ads in parts of the world were *something* dodgy seems to be happening along those lines, eating into my small budget.

Rgds

Damon

4:00 pm on Dec 12, 2005 (gmt 0)

New User

10+ Year Member

joined:Mar 29, 2004
posts:17
votes: 0


Azhar - If you suspect invalid/super-low-quality click activity, check your weblogs, trace back and check out the referring sites and decide for your self it that's where you want to be showing up. If not, modify your campaigns. Pay attention to the quality of traffic coming from search vs content networks, if you participate in both.

My experience has been that ad clicks that smell like fraud to the me (sudden torrents of clicks coming from parked domains, file-sharing sites running adsense, etc.) are still 'valid' clicks to google.

Also, many here may realize ROI from advertising in the content network, but for my business, it's been little more than a place to flush my advertising budget.

If you find a silver bullet to verify genuine ad clicks, please post...

6:21 pm on Dec 12, 2005 (gmt 0)

Senior Member

WebmasterWorld Senior Member 10+ Year Member

joined:Sept 25, 2004
posts:2156
votes: 0


Hi,

Following up from SC above... In one particular case I've had to turn off the "Search Network" (which includes parked domains) whereas the "Content Network" converts fine.

AWA, please ask some higher-up to re-evaluate which parts of the domain-park-shebang are just trash and should be dropped as a possible stain on G's brand... It really is quite ugly what I see coming through some times...

But then, caveat emptor, and at least we have the opportunity and tools to cut out the canker.

Rgds

Damon

3:33 am on Dec 13, 2005 (gmt 0)

Senior Member

WebmasterWorld Senior Member adwordsadvisor is a WebmasterWorld Top Contributor of All Time 10+ Year Member

joined:July 9, 2003
posts:4799
votes: 0


AWA, please ask some higher-up to re-evaluate which parts of the domain-park-shebang are just trash and should be dropped as a possible stain on G's brand... It really is quite ugly what I see coming through some times...

DamonHD, I'll pass your comments on, verbatim, in this weeks Advertiser Feedback Report.

Also, don't hesitate to write AdWords support, be specific about the 'quite ugly' stuff, and ask that it be forwarded to the appropriate team.

----------------------------------------------
An aside to all: I do apologize for my non-presence on this Forum of late. I had virtually no WebmasterWorld access in the week past, and have also simply run out of time for several days running. Just posting now for the first time today, for instance. Access seems fine at present, so hope to post more this week.

11:31 am on Dec 13, 2005 (gmt 0)

Senior Member

WebmasterWorld Senior Member 10+ Year Member

joined:Sept 25, 2004
posts:2156
votes: 0


Hi AWA,

Thanks for the attention (and also remember that I want to be able to automatically turn ads on and off (or budgets up and down) by time of day and day of week which would help somewhat)!

I promise that I do/will point out anything specific that I see that's really egregious, if/when I have enough data points to make a strong case rather than just point to a bad smell...

Rgds

Damon