Results 1 to 1 of 1

Thread: Prevent Selective Pairing and Ratings Deflation&Plateau in Matchmaking

  1. #1

    Lightbulb Prevent Selective Pairing and Ratings Deflation&Plateau in Matchmaking

    Hello dota2 developers, I am the author who recently wrote a few threads about improving the current matchmaking system in order to prevent the number of dota2 players from further decreasing. (Previous threads: http://dev.dota2.com/showthread.php?t=284045; http://dev.dota2.com/showthread.php?t=283257). More specifically, I recently studied some literatures about the ELO rating, which I believe is the currently used rating system in dota2; although I am not a native English speaker, but most of these literatures are written in English so it’s more convenient for me to write my suggestions in English – please pardon any grammar flaws.

    After studying literatures in Wikipedia (https://en.wikipedia.org/wiki/Elo_rating_system) and the references therein, what I realize is that the two most pressing drawbacks in the current matchmaking system is 1) selective pairing and 2) ratings deflation/plateau.

    Selective Pairing includes but not limited to: trying to get matched to opponents who are overrated, trying to avoid getting matched to strong opponents who are rated slightly below them. By exploiting the algorithm in the current dota2 matchmaking systems, there are probably ways to achieve these goals; for example, intentionally decreasing his/her own hero damage, (K+A)/D, gpm, net worth and so on, and thus the system underrates this player. I’ve already discussed this issue thoroughly in my recent thread (http://dev.dota2.com/showthread.php?t=284045). In order to address this issue, one method adapted by Internet Chess Club (ICC) is to enforce “auto-pairing”, where is based on random pairings and “with each win in a row ensuring a statistically much harder opponent who has also won” same number of games in a row. In dota2, this can be translated to using MMR as the sole parameter in matchmaking regardless of their hero damage, (K+A)/D, gpm, or net worth; this can prevent the underrating of certain players. Moreover, if a player is winning in a row, dota2 matchmaking should give the player “a statistically much harder opponent who has also won” same number of games in a row, instead of giving worse teammates to the player.

    Ratings Deflation is defined as the continuous decrease in the average MMR over all dota2 players. Commonly, it is believed that the ratings at the top level (e.g., >7000 MMR) are inflated. However, dota2 players normally enter the ELO system as noobs with a low MMR (let’s say, 2000 MMR) while retire from the ELO system as pros with a high MMR (let’s say, 5000 MMR), and hence the total MMR pool always get a bunch of 2000 MMR players joining while a bunch of 5000 MMR players leaving. Consequently, rating deflation will inevitably occur in the long run.

    Rating Plateau is another outstanding issue in the matchmaking system. In the first figure of this article (http://blog.dota2.com/2013/12/matchmaking/), it seems that dota2 developers are already aware of the fact that, with a plenty of games played, dota2 players’ skills and MMR will eventually reach a plateau. Once this plateau is reached, no matter how many more games they play, their MMR will just roughly stay the same with a tiny fluctuation.


    What are the results of the above three issues? Okay, so Selective Pairing will be exploited by some savvy players and it is unfair for common players, who may not even notice this issue, if at all. However, Rating Deflation and Plateau are actually much bigger issues: it is a common sense that, when players play a plenty of games but their MMR actually decrease, they will just stop playing dota2; similarly, when players play an enormous amount of games but their MMR always stay roughly the same, they will get bored and also stop playing dota2. In either case the rating deflation/plateau will significantly discourage players from playing more ranked dota2 games and if this situation persists, those frustrated players will be highly likely to switch to other video games for their own good.

    Therefore, there remains a demanding challenge, and it is of paramount importance to combat rating deflation/plateau. To this end, my suggestion is to “inject” rating points into the dota2 total MMR pool. This idea has been well accepted for other games and sports in order to maintain satisfying ratings for players over time. Different games and sports have established their own ways of doing it, and what I suggest for dota2 matchmaking is to award 5 points (or 3 points, or any amount of points that dota2 developers deem fit) to players’ MMR each game regardless of their win/loss. Then a player who constantly has 50% win rate will still gain 50 MMR points for sure if s/he play 10 games. If a player is improving quickly and wins 7 games out of 10, then s/he will gain 150 MMR points – good for them. If a player is deteriorating and loses 7 games out of 10, then s/he only lost 50 MMR points – not so bad. One may argue that if this is implemented then the total MMR pool is not conserved anymore. Well, first the total MMR pool isn’t conserved anyway because novices join and experienced players leave; second, if *not* many people are still interested in playing dota2 then who even cares about the conservation of total MMR pool at all. So let’s not worry about the conservation of total MMR pool and first worry about if players are still interested in playing ranked games or not. It can be easily seen that, if playing more games will definitely increase their MMR even if they have a 50% win rate, players will have huge motivations to keep playing dota2 games!
    Last edited by lucifer_fans; 09-02-2017 at 07:31 PM.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •