A comprehensive matchmaking system is an important factor in being able to enjoy playing Dota, regardless of which hero or role you prefer. Today’s update includes some engineering changes to the core matchmaking system to allow for bigger changes and better analytical capabilities. Over the next year, we’ll be spending more time focusing on various aspects of matchmaking such as intra-team balance, player conduct, new player experience, abusive behaviors, friend and teamplay aspects, high mmr matchmaking dynamics, and other issues in an effort to make the overall experience of playing Dota more fun for players of all levels. In this blog post we decided to drill deeper into the changes in today’s update and explain our thoughts.
Matchmaking involves various tradeoffs and our goal is to find the best balance between all the different factors. In order to help make progress on finding the right set of tradeoffs, today’s update includes a couple different experimental changes that will last until the end of the season. Close to the end of the season we’ll do a direct call for feedback on how the changes have worked out and use that feedback, along with data we are gathering in the meantime, to help inform the direction for the next season. We’ll also be working towards other features, such as better detection of smurfing and other abusive behaviors as the year progresses.
The first of these experiments is removing the concept of separate Solo and Party MMRs. We expect this to be the most controversial component in today’s update, as we’ve heard fair criticisms of it in the past. We can’t say that we are fully confident that this will end up being a good change either, so we’ll want to wait until the end of the season when we gather feedback and data on how it has played out after players have had a chance to absorb it. There are two major aspects with this change: 1) the teamplay and social aspect and 2) the value and correctness of the MMR value.
For the teamplay aspect: We think it is really crucial for you to not have a disincentive to play with your friends. The game currently overly emphasizes playing solo and establishes a strong social reward mechanism for this, which causes many players to prefer playing alone than with friends. We believe that bias has over time caused more negativity and unhappiness when playing Dota. Furthermore, Dota is a very teamplay heavy game and we want to consider that aspect a bit more strongly in the hierarchy of matchmaking values. We’ve considered other approaches to the friend and teamplay incentives, but they tended to do a poor job at making it feel actually rewarding to play with friends and only papered over the issue.
For the correctness of the MMR aspect: We recognize that there is a tradeoff here on the mmr data quality if the match has solo mmr numbers with party mmr numbers, however we feel that the impact of that data noise is much smaller than even we initially considered it to be early on and generally with how it is discussed online. Most of the issues related to this we think we can solve with better algorithms. Another common case is playing with a friend who has a higher mmr than you, expecting that it will cause your mmr to go higher. While that has been at least partly true in previous matchmaking system iterations, our most recent version does a fairly good job of addressing this. We have enough data to form matches that cause you to have an even chance to win by carefully choosing the opponents that have a similar makeup. At a base level, we think mathematically we are able to have high confidence that playing with a friend will not have a material effect on your average mmr value with a properly balanced matchmaker. We’ve done some testing to make sure this is true in advance of this update, however we know it’s likely there is more work to be done here and we’ll continue working on this aspect in the background during this experimental period.
Through our investigation into the topic of matchmaking volatility and how parties affect the game, we’ve noticed something that fans likely already know: players have very different comfort levels when playing core roles versus support roles. Some players perform really poorly when they play a role they are not used to, and in turn introduce a lot of volatility in the system for everyone by causing your MMR number to drift away from your actual skill level as well as causing imbalanced games.
So because of that, and because of our abiding love of having multiple MMR numbers, we are adding a new concept of Core and Support MMR numbers. In order to achieve this, we need to know what role you want to play in advance of the match being formed. So for this experimental update, we are moving the Ranked Roles feature to the base Ranked matchmaking and expanding on its capabilities. Now when you matchmake, you will have the option of selecting roles from position 1 through 5 (Safe Lane Core through Hard Support), including multiple selections. So if you like playing Mid or Offlane, you can select those two, or if you like playing Hard Support and Mid, you’ll be able to do that as well. You will be matched based on either your Core or Support MMR.
Each MMR type will also have its own leaderboard, and for the purposes of this short experimental period the existing rank will be applied as the initial rank for both Core and Support. If we end up keeping this feature in the future, there will be a separate calibration phase for each one.
These experimental changes will remain active for the rest of this season for you to try out. Sometime after The International concludes, we’ll issue a call for community feedback to help guide the next steps in our efforts to increase matchmaking quality for everyone. We hope to have your support and patience through this experiment as well as subsequent changes towards the path of making a better overall matchmaking system.
While we were initially going to release this change as part of the Summer Scrub update, the amount of significant underlying code changes required for this update introduces a lot of potential instability. So we are splitting the Summer Scrub update into two parts, this first part for a matchmaking rework, and the second part soon after for bug fixes and a few quality of life additions. We are expecting that we’ll have a lot of matchmaking related bugs to work through today, so please let us know if you run into any issues and make sure to include the matchid in your reports.
For the matters, we've service that'll provides you with Valve's latest movement first hand. You can check it out for more details about our new and improved calibration service here: Link