What don't we just use a spam filter for comments?

in steemit •  7 days ago

We all complain about the comment spammers. I know this is disgusting. However, with the current state of machine learning, implementing spam classifiers are so easy.

road-sign-464657_640.png

Source


While I was reading Compilation of "Shit Comments" on Steemit Vol.2, I had the idea of classifying these comment patterns and creating a REST api for comments if they're spam or not.

Since spam comment patterns are generally very clear, it should work with few false-positives. Then interfaces can implement this API to ignore this comments by the preference of the user. Or, we can build browsers extensions doing that.

That sounds yummy. I will implement this right away. It shouldn't be too hard to train the sample data and analyze the results.

Thoughts?

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!
Sort Order:  trending

Sounds good although, I'm a noob when it comes to coding 😅

·

Then you can stay safe on the user side :)

Sounds great! Are you going to use Naive Bayes or some kind of neural network? Or maybe different approach?

That would be great if you can handle this. What they are truely gain from comment spamming. I don't understand that part :D

·

I read somewhere on steemit that these spam comments usually get upvoted around 25% of the time. I guess that’s actually a decent amount... the up votes are probably coming from noobs like me who are just happy in the beginning to see any comments at all. I know I upvoted them in the beginning. 🙈

I think this platform has enormous potential. The rewarding system promotes honest engagement in spite of passing through incentivized engagement...albeit it has the flaw of making spammers such a frustrating reality.
I am still a newbie, so it hasn't got to me yet, but i am starting to feel those comments too. Even more so when i see the few people that i follow complain about it and even getting bitter.

The browser extension idea sounds the most practical, doesn't it?

You got a 2.02% upvote from @buildawhale courtesy of @emrebeyler!
If you believe this post is spam or abuse, please report it to our Discord #abuse channel.

If you want to support our Curation Digest or our Spam & Abuse prevention efforts, please vote @themarkymark as witness.

I was gonna do that! Didn't get around to it yet but I'd love an extension like this one and that was high on my TODO, can't wait for your implementation. I think in this case a kNN neural network would probably do the trick, since you only have 2 outputs and a 2d plane?

·

I would go with the Naive Bayes classifier. Simple to implement and fast to run.

·
·

To be honest, I will use plug and play libs. I should prove it first, then I can optimize it.

·
·
·

Still, most libs like Tensorflow allow you to choose the Classifier, you should test which works best!

This post has received a 3.69 % upvote from @booster thanks to: @emrebeyler.