When rating players using Elo system, sometimes we have additional associated player type that may affect the game result, this additional player type information in some cases may help predicting game result, e.g. players with specific type may perform better playing against players with some other specific type.
Consider the game StarCraft: Brood War, there are usually pre-determined player type (race) of either Protoss (P), Terran (T), or Zerg (Z) for each player in a one-versus-one matchup. Each game is either PvT, TvZ, ZvP or mirrors like PvP, TvT, or ZvZ in race matchup. Given game result data with player type (race), we can compute the total Elo ranking by discarding the player type data, or have individual race matchup Elo ranking like PvT, TvZ, etc.
When predicting game results, using individual race matchup Elo ranking takes into account of the player type, however for each player it discards other types of matchup data of a given player, where the strength maybe highly correlated. These other types of matchup data may provide good information to improve on the vanilla Elo ratings if the specific race matchup data are sparse. Are there extensions of Elo system that takes into account this information, thus combine both information?
