I was just wondering if a Riot employee could tell me how it is determined for a player to be put on chat restriction. I know it is an automated system but how does it process a report to tell if a player is toxic or not? Just want a clearer answer because my friends have been getting them recently but they are rarely toxic.
I've posted a lot about how this all works recently. Rather than write it up again, let me offer a couple resources for you from the old TBI forums:
This thread primarily relates to the Tribunal, which isn't active at the moment, but also includes a fair amount of information about the report system and how it works. This is key, because with the exception of Leaverbuster, all of the behavioral systems currently in place rely on the report system. Here is a specifically relevant quote:
“How many reports does it take to get to Tribunal?”
They are NEVER going to give us exact numbers because stupid people would try to game the system. Here’s what we know:
- A case is not triggered by raw reports. That is, there isn’t a magic number that gets you into the Tribunal.
- Instead, a case is triggered by report density. Lyte has said that most people who trigger a case have been reported in at least 15-25% of their games. We don’t know the time frame except that they do a purge at the end of each ranked season.
- It would be logical for the time frame to be relative to the density of reports. That is, someone who is reported in 75% of their recent games would trigger a case in days, maybe hours, while someone who was running close to the 15% lower bound might take weeks or months.
This thread more specifically relates to the current behavioral systems, specifically those which govern the waves of escalated bans, and the chat restrictions. Again, a specifically relevant quote:
"How did they determine that I was toxic?"
As part of the development of Team Builder, Team PB&J built "sophisticated language models to determine the quality of communication." Lyte has mentioned on several occasions that they could - and probably would - use those models to detect toxic chat. Given Lyte's penchant for accuracy, it seems likely that the detection system was thoroughly vetted before release.
From WookieeCookie, link
- The chat restrictions are not based off of the last game you played but rather an investigation of a series of games you've been reported in. Just a few of the things we look at: frequency of games played, frequency of reports received, # of unique players reporting you, types of reports received and so on.
From Lyte: link
- Secondly, we have a system that's been analyzing millions of chat logs, and become quite good at detecting whether a conversation in a game is negative or positive. I presented some of this work at the previous Game Developer's Conference. It's an interesting system because the more chat logs you feed the system, the more accurate it becomes at determining whether a player communicates more positively or negatively than the average player in League of Legends.
Keep in mind that Riot will NEVER release details about this detection system, as people would use it to skirt the system.
There's plenty more really great information in both, and I highly recommend giving them a look.
