Machine learning methods and neural networks should be applied to 4chan moderation. Recent advances in deep learning would help a lot I think.
What about a system that learns from every mod/janitor deleted post?
You can't just train on deleted posts, you need to give it both deleted and non-deleted posts so it can learn to distinguish the two. If there's any inconsistency (e.g. a mod/janitor deleted something they shouldn't, or missed deleting something they should have) then you'll get inconsistent behavior from the trained machine too.
I suspect that most deleted posts are for NSFW images in SFW boards. Unless the machine has been trained to identify these too, it won't be helping much.
>>53711641
>You can't just train on deleted posts, you need to give it both deleted and non-deleted posts so it can learn to distinguish the two.
Well every non deleted post is a good post, so what's your point?
>If there's any inconsistency (e.g. a mod/janitor deleted something they shouldn't, or missed deleting something they should have) then you'll get inconsistent behavior from the trained machine too
That will matter less and less as the training continues. That's what deep neural networks are good at, you think large image training sets don't contain errors?
>I suspect that most deleted posts are for NSFW images in SFW boards. Unless the machine has been trained to identify these too, it won't be helping much.
Of course the training would include nsfw images recognition. Again, that's what deep learning is good at.
What do you mean "learn from a post"?
How will you feed text to a neural network?
With images its easy, since you have actual pixels with brightness and color and stuff.
You haven't been working much with neural networks have you OP?
So basically it would be a cuckold simulator.
>>53711794
Images make most of unwanted content on 4chan, so maybe that's not a problem and images recognition only will be enough.
Also people have been experimenting with nns and text for a long time now.
>>53711845
>people have been experimenting with nns and text for a long time now.
i'd love to see some projects/citation
>>53712203
Really?
http://arxiv.org/abs/1603.07141v1
http://arxiv.org/abs/1603.06127v1
http://arxiv.org/abs/1603.03915v1
http://arxiv.org/abs/1603.01547v1
http://arxiv.org/abs/1602.07480v1
http://arxiv.org/abs/1602.07475v1
http://arxiv.org/abs/1505.04342v5
http://arxiv.org/abs/1602.06979v1
http://arxiv.org/abs/1602.06797v1
http://arxiv.org/abs/1602.06359v1
And these are only very recent ones.
