Behavioral Systems: October 2020 Update

Welcome back to our October edition of Behavioral Systems updates. We have some new topics for everyone this month, but let’s take a look at how we are currently tracking towards our goals, and what’s coming through the pipeline next.

Goals and Status Update

We started the journey a few months ago with the following goals specifically targeting Intent to Lose and AFK behaviors:

• Mitigate the impact of behaviors when they occur in game.
  • Early Surrender (In Progress)
  • LP Mitigation: Providing LP consolation in ranked games for verified AFKs and intentional feeding (In Progress)
    
    
• Deterring this behavior from occurring in the first place.
  • Penalty and queue lockout increases for verified AFKs (Launched 10.19)
    • Bumped up AFK penalties applied by 10% globally
    • Penalty increase correlates to a decrease in overall AFKs region to region
  • Detection improvements for AFK behaviors (Launched 10.20)
  • Detection improvements for Intent to Lose (In Progress - estimated 10.21)
  • Detection accuracy updates for all behaviors (In Progress - more below!)

In 10.21, you'll see an increase in intentional feeding and griefing player detection. We've taken a pass over our existing detection that relied on players repeatedly feeding and accumulating many deaths. Our 10.21 updates look at specific details of each death to flag those that appear to be intentional. This allows us to detect players with only a few deaths while still being able to separate intentional feeders from players who are just having a bad game. 

Our goal with each detection update is to hit at least one of the following without regressing on the others:

• Detect more offenders than before
• Detecting offenders sooner
• Reducing collateral damage from falsely detecting innocent players

Our current focus is still on penalizing severe offenders after just one game, but some of the nuanced behaviors may take a few games before we have enough confidence to issue a penalty. In the next update, we’ll be looking at moving those behaviors that take multiple games to penalize with confidence into behaviors we’re feeling confident penalizing after just one game. Longer term, we'll be exploring ways of dealing with intentional feeders as the game is in progress.

Hot Topic: Reporting Confidence and Accuracy Updates and You

In our August update two months back, we referenced our manual review of hundreds of games and from doing so, we’ve learned a ton. Jumping right in, our biggest takeaways were:

• Player reports are more efficient at identifying the symptoms of more subtle/nuanced game-ruining behavior than us trying to define those symptoms through theoretical conversations.

• Not all reports are created equally. There are players who are great at consistently recognizing disruptive behavior and others who spam the report button every match.

• We're pretty confident in the progress we're making with improvements to our single game detection systems, but we learned that approaching games one at a time doesn't capture the full picture of how players engage in various behaviors across games—even if each individual game is accurately assessed

With this in mind, we can’t wait to introduce y'all to our newest detection system that we have been iterating on for months. Using the dark magics of big data, we have completed our first iteration of our new post game behavioral awareness system. 

The system works by measuring and controlling for variations in player behaviors over large periods of time. This gives us the ability to distinguish between games that just didn’t go your way because you were stomped in lane versus games where your teammate ran circles around your nexus for 20 minutes. By leveraging a longer player history, we unlock the ability to better inform decisions on when to issue penalties and when to back off.

Players can expect that finalized system to do the following:

• Identify players that consistently engage in disruptive behavior for harsher penalties than our current single game detection system. The more frequently one engages in disruptive behavior, the harsher the penalty. This is already in place for AFK and Leaver penalties, but we'll be expanding it out to other disruptive behaviors.

• Distinguish between the players that are consistently great at recognizing other players' disruptiveness from those who just spam the report button every match.

• Allow disruptive players the opportunity for behavioral reformation. One of the benefits of our new system is we can track behavioral shifts over time. We want to allow players who reform the chance to come back, while still being able to effectively prevent bad actors from ruining games.

We’ve tested this system on players intentionally feeding, griefing, and abandoning games and are feeling very optimistic. Our early tests indicate that we can simultaneously increase both the number of detections as well as our confidence in their quality.

In the upcoming months we'll be tuning this system to increase the confidence of our detections, expanding the breadth of behaviors we can track, and going live.