How accurate could you make an algorithm that predicts the number

>>8383090
This is a perfect application of machine learning.

Equip a program with only the tools to browse 4chan. Have it request the catalog multiple times to keep track of posts that come and and go and so that it can create its own internal catalog of alive threads.

Then through another thread (in the multi threading sense of the word) it should browse the links it has stored in its list (which would be the threads) to see how many replies they get while it keeps measure of time and it stayed alive and replies.

Obviously also give it tools to record the words used in the post and the general structure used.

After about a month it would be perfect, assuming that we would get meme-frozen by which I mean a state where we no longer create new memes but instead just keep reusing pre-existing memes. Kinda like 9gag.

One important variable that has a big impact on the number of replies and which is very easy to determine is the time and weekday at which the post is submitted.

I think one would be able to make reasonable guesses when considering aforementioned factor plus applying statistical methods on a comprehensive dataset of "content - number of replies".

>>8383090
Wildly inaccurate is my guess.

I'd hit it with the meme hammer and do some variable sequence length regression analysis by feeding an LSTM neural network character level posts and training it to output the number of replies.

ez

>>8383103
>>8383110
>>8383112
nice

would it be possible for it to recognize memes using image processing?

>>8383103
>After about a month it would be perfect, assuming that we would get meme-frozen by which I mean a state where we no longer create new memes but instead just keep reusing pre-existing memes. Kinda like 9gag.
nithmare fuel

File: extremely rare devil pepe do not copy without permission.gif (2MB, 425x481px) Image search: [Google]

2MB, 425x481px

>tfw creating an evolutionary neural net program that posts on 4chan and iterates itself based on how many replies it gets, producing ever more effective baits and shitposts
who /devilish/ here?

Who want's to start a open source 4chan shitposting neural network with me? We could use several Raspberry Pi's harvesting data for a few weeks and after that building a shitty neural network with tensorflow. Who's in?

File: this place.jpg (9KB, 225x225px) Image search: [Google]

9KB, 225x225px

>>8383156

fuck you DeepMind

>>8383131
yes

>>8383161
Just do it yourself fag.We all know you are trying to make us do your senior thesis.

>>8383172
Assuming I will write a script for collecting the necessary data for such an endeavor, would you be willing to run it on your pc/raspberry pi? I'm a college poorfag only owing one filthy little raspberry pi and a laptop.

>>8383156
Dear lord shits going to go so badly. Imagine a world where you can go out and torrent an iterative shitposting machine. At a certain point all posts will just be shitposts that have been calibrated to be most effective turning the bots on each other in a vicious cycle of competition. Soon an arms race begins to create a more effective shitpost. People make bots that respond to posts to get the most responses per thread, the bots begin to work in tandem where the OP bots will create threads that the responder bots can easily drum up conversations in. Meanwhile other bots will form parasitic relationships killing threads faster but with a higher amount of responses to the responder bots meaning other bots will have to start making defense mechanism against the Troll variants. Time passes and soon all humans are removed as they are unable to get any responses in competition with the interesting things the bots are saying, but no one notices because it looks just like before simply stagnant.

File: it will be alright.jpg (6KB, 279x180px) Image search: [Google]

6KB, 279x180px

>>8383175

how can you be so sure it hasn't happened already? :6)

[spoiler]look behind you[/spoiler]

>>8383174
Why do you even need to run it on several computers?

Just have it multithread but even that is overkill. /sci/ is the slowest board so just have it explore /sci/, it won't miss any detail if it takes a while looking through posts.

File: Untitled_1.jpg (191KB, 1280x720px) Image search: [Google]

191KB, 1280x720px

>>8383139
Ironic

>>8383212
because the shitpost/time ratio on boards like /b is fucking high + the connection speed is the bottle neck... I'll write it anyways and post the finished script here on /sci, maby tomorrow