Introducing: Performance-Based Matchmaking

As mentioned in the BlizzCon 2017 opening ceremony, and during our “What’s Next” panel, we will soon be introducing performance-based matchmaking to Heroes of the Storm. With this change, your matchmaking rating (MMR) adjustments will be influenced based on your individual performance in the match instead of being purely based on whether you won or lost. We’ve broken down some details of the new system below, along with a short FAQ. Read on for details:

What’s Changing and Why:

In our current system, an individual player’s Matchmaking Rating (MMR) is primarily determined by the amount of games that player wins versus how many the player loses. When you win a game, your MMR goes up, while the opposite is true for when you lose. In addition to that basic rule, your MMR will go up more, or less, based on the comparative skill of both yours and the opposite team. If the enemy team has a higher MMR than your team, you’ll gain more points for defeating them and lose less if they win. The opposite applies to enemy teams that have a lower MMR. However, the team-focused nature of Heroes presents challenges when using this system to determine an individual player’s matchmaking rating, since any single player is only 1 part of the 5-man team that won or lost the match. The system works since, all things being equal, a player will win more games than they lose over the long run if their skill is higher than other players at the same rank. However, unless the difference is significant, it can take a significant number of games before the difference in wins vs losses adjusts the player’s MMR enough to match their true skill level.

With performance-based matchmaking, we’re looking to change that. Let’s take a look at how we’re planning to make matchmaking more accurate and better able to recognize individual skill in a team-based game:

How it Works:

Once we introduce performance-based matchmaking in the near future, our system will look at how well the individuals in a match are performing and use this to aid in calculating their MMR adjustments for the match. The system will apply context to these stats by looking at the Hero you’re playing, the Battleground you’re on, the game mode, and what region you’re playing in so it can make fair comparisons against other similarly skilled players under the same conditions. Our goal with this system is not to define a specific playstyle for each hero, since we are never manually setting which stats are considered important, but rather to let our machine-learning driven system define importance and weighting for stats based on millions of games of data. By looking at these stats across the entire player base, the system can see which stats are most important for effective play and create a data-driven model of what the most highly-skilled players are doing in any given situation.

What differentiates a highly skilled player on a given Hero isn’t always obvious, though. For example:

When looking at a Hero like Illidan, it’s pretty straight forward. A good Illidan does a fair amount of damage, soaks a lot of experience, doesn’t die very often, and captures a lot of Mercenary camps. Illidan has a fairly high skill ceiling, though, so we expect there to be a sizeable difference in all of those stats between a highly-skilled Illidan and an average one.
However, for a hero like Kerrigan, the difference in raw stats like damage and experience between a high skill and average skill player isn’t quite as dramatic. Instead, the biggest thing that sets apart a highly skilled Kerrigan is how effective they are in landing crowd control effects, so the system would put a larger emphasis on this stat for Kerrigan players. She’s still being evaluated on all the same stats as every other hero, though, so a Kerrigan who spends the game only attempting to land stuns would not necessarily gain the same amount of MMR, as they would need to also contribute to the game in a meaningful way on top of landing lots of CC.

The system can then use an individual player’s performance to adjust how much MMR the player gains, or loses, for a game. If they’re performing beyond expectations for their current MMR, the system will grant more MMR when they win, subtract less when they lose, and they’ll more quickly arrive at the proper MMR for their skill. The opposite is also true if they’re not performing up to expectations.

FAQ

What is MMR?
- MMR stands for Matchmaking Rating. In order for the matchmaker to place similarly skilled players together, it assigns everyone a matchmaking rating that indicates how skilled the system thinks the player is. MMR is a behind-the-scenes stat that is only used for matchmaking.
What if a Hero is updated and the “right” way to play them changes?
- The system is fully dynamic and continually updating itself, so it automatically adjusts as balance changes, the meta shifts, and players find creative, new ways to use a hero.
If I perform really well, can I gain MMR while losing the game with this new system?
- No. Winning or losing is still the primary factor in whether you gain or lose MMR so you still only gain MMR on a win and lose it on a loss. The performance-based adjustments just affect how much MMR you gain or lose.
Couldn’t you abuse the system by ignoring everything except those few crucial weighted stats?
- While some stats are weighted higher than others to reflect their importance in that situation, every stat is evaluated as part of the performance calculation. If you’re focusing entirely on a few stats at the expense of others, your overall performance metric is likely to be lower. Also, you still need to win the game to gain MMR. If you’re maintaining all stats at their normal levels for your level of play, excelling at a few key stats, and winning the game, that means you’re actually doing what’s required to be the most efficient, effective version of your current hero. Good job!
If I’m losing a game, my stats are usually lower already. How does the system avoid double penalizing me for both losing and having lower stats?
- The comparisons to determine how well you performed are different for winning games versus losing games.
Will this cause players to intentionally make games take longer so they can pad their stats?
- All stat comparisons take game time into account so there’s no advantage for a game that goes long and no disadvantage for one that wraps up quickly.
Who determines which stats are important for any particular situation?
- The community does by playing the game. The system doesn’t have any preconceptions about which stats are important. Instead, it is measuring how players are playing in particular situations in order to determine which stats are most important to highly skilled play.
How does the system work for a new hero or after a hero is reworked?
- The performance-based aspect of the system will be disabled initially when a new hero launches or after a hero receives a major rework to allow the system to gather the data it needs to make performance comparisons. During this period, the system will effectively work like it used to with MMR adjustment based on win or loss.
This system is affecting my MMR, but how will this translate into my rank in Hero League or Team League?
- We’ll be introducing a Performance Adjustment to the amount of rank points you win or lose after a game which mirrors the performance adjustment that was applied to your MMR. This will eventually replace the current Personal Rank Adjustment.
Will this help with smurfing?
- We believe this system will go a long way to curbing the issues associated with smurfing since a player who creates a new account to intentionally best players of a lower skill will more quickly find themselves facing players of similar skill.
When is this coming to the game?
- Our current plan is to implement performance-based matchmaking in December alongside our next season roll, but we will be sure to update you if anything changes.
Awesome. Anything else I should know?
- The system has also shown promising potential to help detect griefing in games since it can more easily determine when someone is intentionally playing poorly. We’ve already used it to confirm reports of players who are doing things like intentionally feeding and hope to be able to automate that detection in the future so problematic players can be dealt with more quickly.