The channel commercial false quantity pretends the real user: the developer asks the way to live

Intermediary transaction http://www.aliyun.com/zixun/aggregation/6858.html ">seo diagnose Taobao guest cloud host technology Hall

Entrepreneurs are very bitter turtle, suffer from creativity, suffer from development, suffer from user experience.

Entrepreneurs are very bitter turtle, even if the product is done, do not promote also difficult to obtain users.

Entrepreneurs are very bitter turtle, in the absence of good channel business, spend money to do the promotion, spend time to do change, but the product is still no improvement.

Look at the readme of a similarly bitter developer Choodong shoes.

I do data mining technology background, and engaged in the promotion of three years, app paid to promote the phenomenon of fraud has not uncommon,

This time at last, I have seen the false, and have not seen so much nonsense.

Tens of thousands of yuan to promote the cost, get tens of thousands of activation, only 7 real users, of course, do not rule out these 7 "real" is this pit Dad channel testers, products, operations, business parents must be careful:

My company for a long time to do app change and pay promotion, the recent on-line set of intelligent algorithm used to identify the false amount, found a "channel a" (The channel real name can be asked Zhao Dong shoes in private) serious cheat brush false amount. Take a look at the last week of the channel a data (note: Due to the product confidential data, the specific values are hidden, only the overall trend of performance.) All charts are as follows:

• Registration Rate

• Retention rate

• Startup times

(1) Registration rate

(2) One retention user

(3) Starting times (average per person daily)

There is no obvious anomaly in all the data, and it is a "high quality channel" from the data. Unwilling to?!?! More depth data (last week):

• User Churn speed

• User visit interval

• Hardcore user ratio

(1) User churn speed

(2) User return visit interval

(3) Hardcore user ratio

Commonly used data are checked, there is no conclusive exception! Well, whether you want to accept it or not, the reality is:

Traditional data indicators do not distinguish between false channels.

How to do, there is no data can make fake nowhere to hide?

It can be imagined that the fake amount of robot is not a real user after all, they complete the "designated" Operation task, no longer because see beauty on her head, will not play a little bit more fun. So take a look at the comparison of user-length curves:

Channel A is used for a long period of time below the overall average (other channels), and does not form periodically with the date of the regular oscillation. Finally is away, completely can determine is the robot brush false amount!

Robot brush false amount, should be able to find characteristics. The previous practice is to compare the IMEI number, but now the brush robot has been able to fake fake IMEI.

But still can look for the robot characteristic according to the IMEI, the normal channel's IMEI number is very few to repeat, and the repetition probability does not increase with the time delay, but the cheating channel's IMEI repetition rate is often high, and grows with the time unceasingly.

Can actually IMEI data still can't see anomaly, need to think of another way. Different users of mobile phones, are used by different communications operators (such as Mobile, Unicom, telecom) phone cards, through the statistical operators have the hope to find out the characteristics of the robot. In order to do analysis, sampled a handful of users, extracted their carrier information, statistical results show:

It seems that through the operator's case is not the analysis of the characteristics of the robot. And I also sampled some of the user IP address, whether IP distribution or IP across the province and the probability of the same as the normal user.

A bloody reality lay before us:

The false user's simulation degree is more and more high, it is more and more difficult to identify it, and extracting the false feature is almost impossible.

Of course, there is no bitter turtle to walk the road, eventually our company still completed the evidence.

The sample extracts the list of apps installed by some Android users in the channel and finds that many users have installed apps that are identical. Dig this trick again, the result learned:

A channel, a total of 4 sets of robots, the number of robots per group, the same group of robots installed in the same app, and no increase.

Finally, a channel of the robot pulled out, but because of the laws and regulations of the problem, we can not put the channel how, can only not give money, and then tell false news to know friends.

Moreover, the result of the event is not as happy as the fairy tale, the bitter turtle we met the more invincible robot brush, and all of the above methods of judgment all fail. It's villains, outsmart.

So far, with my company's channels, nearly 30% detected a brush false amount of cheating, from the 2010 the simplest of the mentally retarded irrigation, to 2011, the emergence of intelligent robots, will begin to simulate the real user registration, but also to return to the 2012, the robot has been able to achieve personalized high simulation.

And then think about it, such a channel can provide such a high imitation of fake users, the traditional data indicators are invalid, so that ordinary developers can not distinguish, the biggest reason is because of the driving force, the cost of tens of tens of thousands of of millions of active but only 7 real users, this black-hearted money earned too sharp, this is almost 0 of the cost of profit!

And now there are not many companies in the search for cheating, the low risk of cheating, and there is no relevant regulatory constraints, fraud has become a matter of course.

In addition, as far as I know, do the fake volume of the activation of CPA, the unit price only between 1 to 7 cents, much lower than the true amount, the middle of a huge difference in the gap has brought huge rebate space. The person responsible for the promotion may have found cheating, but it is also possible to be motivated by kickbacks or performance pressures.