Boards›Player Behavior

toxic player detection and an automated response

Bodynox·12/14/2018, 12:24:10 AM·2 votes·1,685 views

Hello, I am a graduate student in a data science program, for my thesis, I proposed a machine learning problem that I would like to attempt to solve. The problem is that cyberbullying aka toxic players is a thing that happens in games such as league of legends and with a reporting system such as the one currently in place where it requires players to report other players, we know the turn out it actually very low. my solution was to have a machine learning algorithm in place that could detect cyberbullies based on what they type in chat. I would like to have some real-world examples of these logs so I could use them as a learning set and test set for my algorithm, and I wonder if people would be more friendly (less toxic) if they knew they had a constant big brother watching over them, but also does anyone know a good way to get large quantities of chat logs from games that fall under the lol term and conditions?

12 Comments

rujitra12/14/2018, 12:32:00 AM5 votes

If nobody reports something, why should a player be punished? As an example, if everyone is making jokes and using a "no no" word, but they're either all friends or all okay with it, why should a player be punished just because a machine is offended?

And no, I doubt Riot will give you chat logs. They already use machine learning in their systems and there's very little incentive for you to do research on it. I get you're looking for a thesis that you'd enjoy researching, but machine learning in video game punishment systems is something I think you should avoid. You will have a very hard time doing research on it that has not already been done and published, which will make your thesis very, very hard to do as you can't generally duplicate prior research without very good reason.

My suggestion to you is to choose another topic for your thesis.

Vreivai12/14/2018, 12:27:32 AM5 votes

I don't need a machine being offended on my behalf, thanks. If I care, I'll report.

usul120212/14/2018, 12:33:12 AM3 votes

The IFS is already using a machine learning algorithm though..... (and riot is okay with chat that isn't reported. In groups of 5 I know my friend group is certainly a but less than honorable, and that's okay because we're all in on it)

R107 Games12/14/2018, 12:32:28 AM2 votes

Please no.

Chat is moderated harshly enough. We don't need it to start punishing for everything and make players analyze what they type before pressing enter.

Telephone Booth12/14/2018, 4:04:47 AM1 votes

Yikes. This type of shit is scary to me for some reason. Its like youre contributing to a future dystopian world where big brother makes sure you are a happy little sheep spreading positivity while the elite hoard all the resources and arrests anyone who says anything bad about society. Lol i watch too many movies. But the tone of you seeming to WANT a big brother, is indeed fucking terrifying, because eventually you will be in the majority.

FOR JUSTICE12/14/2018, 1:05:07 AM1 votes

IFS is already a learning AI. Also if the chat isnt reported it wont be scanned. I can imagine it looks for common factors to guess what is toxic.

Icy Hot Shoto12/14/2018, 1:08:51 AM1 votes

Except it wouldn't work.

It would be exactly like when Riot gave infinite chat bans. Instead of being toxic in chat, they'd be toxic in other ways. Either by inting, trolling, or both.

mlm olo mlm12/14/2018, 1:28:31 AM1 votes

Just my opinion....

The problem isn't actually the words. When a person becomes angry, they are prone to react instinctively. It's kind of like shouting when you accidentally bang your toe.

So, an algorithm would be able to identify negativity, but it can not prevent negativity.