"Neural network" to identify pests on site?

Collects, analyzes and issues a ban if the user believes that he is a fraud, or harm the website. What would you recommend? Who have done similar or worked with these? Library? Source?

Let's say a user is spamming on the form, the blog, on the wall, the neural network is the thing firing and issues a ban. Or the mother of the participants of the forum in your language, etc. Manually monitor all unrealistic, even if you put filters, filters will be many, and all to put unrealistic.

Probation will coeficient of viditelnosti, say 100% eto finally, the pest direct avid. Network reduces or increases coefficient. Let's say the user nasami on the forum, and the severity 20%, the network provides the Mut on the forum, blog, etc., the user can't comment. The user starts to write a PM and send spam + 50%, blocking some actions. At 90% to do anything to it, just go to the website and write in support. If starts to do what podatki, like to parse a website, create a burden and an obstacle to his health, it's all the same it is necessary to monitor, manually understand what functions, and this is just a basic example. However, after some time, coefficient begins to fall, if the user deletes his spam, and starts for example to do fumbling the posts in the connected Twitter account on our website. It brings profit, the network monitors it, so it is necessary to encourage the user, reducing COEF sabotage, and give some message, or a bonus to it.d.
June 5th 19 at 21:54
2 answers
June 5th 19 at 21:56
To manually monitor all unrealistic, even if you put filters, filters will be many, and all to put unrealistic.

And what your neural network ? Button - make beautiful ?
The algorithm itself learns, gathers info, analyzes. There are no clear boundaries what is bad and what is good on the website, it would be insane to set all the parameters, although this is probably already going to AI. In General, an interesting question for me, I don't rummage through neurons want to hear the advice and thoughts of others. - jaycee commented on June 5th 19 at 21:59
, a neural network, it is not a Swiss army knife, and the specific implementation. For example Google have introduced 1000 people to buy an elephant, then when you enter the buy may 1001ый also wants it, so put it in the tips first.
What you need, 99% will not be in the nature. There will be some ready-made anti-spam modules and just simply the price for them will be directly proportional to their functionality. - tomas.Remp commented on June 5th 19 at 22:02
The network itself is trained. The network is trained. For that, she would have learned that it is necessary, first, she had to slip a couple of hundred (well at least tens) of thousands(!!!) examples that partition manually (!!!). Do you have them? Can you imagine how such a training sample to collect? - Katelin_Greenholt77 commented on June 5th 19 at 22:05
no - jaycee commented on June 5th 19 at 22:08
June 5th 19 at 21:58
No spam, no neuronox, but perhaps useful
https://habr.com/company/southbridge/blog/255343/

Find more questions by tags PHPNeural networks