Machine Learning System

1. About

 

The Machine Learning System is a system that collects all player-made reports and learns (hence the name) of what and why players consider punishable while playing League of Legends [31], and feeds them to the Instant Feedback System.

 

2. Implementation

 

Since its launch, the machine learning system has analyzed, and keeps analyzing, several pieces of data obtained through player-submitted reports and honours [262, 464], which contain info on how to classify a specific behaviour as positive or negative, and even on how to discern context, languages spoken and language or game-specific slang terms [66, 147]. The peculiarity is that every piece of this data differs from region to region, so that the Instant Feedback System is tuned to local cultural quirks [66, 146], and it’s always up-to-date thanks to the ever-learning Machine Learning System [147], which can also easily spot changes in harassing behaviour, as long as players keep reporting them [386].

Report accuracy is another factor that the machine learning uses to discern toxic patterns when unsure about the context. In some cases, it can lead to punishments against the reported player, if the reports were made by players with a high accuracy [416]. Accuracy is an important factor in deciding a player’s report weight; it can decrease with false reports, and increase with justified reports [183], but never with sheer quantity [418].

 

3. Miscellaneous

 

• At some point, there were plans to improve the Machine Learning System in a way that it would allow players to dodge champ select without getting penalized, but only after reporting suspected trolls accurately [133].
• The ability to identify positive behaviour was implemented to allow Riot to deliver positive feedback and a small prize instantly to the players identified as positive [178], which is a plan that, to this day, hasn’t been put to practice yet. It’s also able to recognize when players are trading honours, and make them meaningless [464].
• The machine learning engine took months of research to be able to recognize more than fifteen languages, while being able to adapt to changes [345, 473]. The results of this research were incorporated into the new Tribunal [473].
• Machine learning has been stated to be “not necessary”, but the results are worth the development costs [286].

 

4. Sources

31 http://hwcdn.libsyn.com/p/d/e/d/dedcb49677a8dcd1/PoGCast_Episode_4_-_Toxic_Behavior.mp3?c_id=9204818&expiration=1451594700&hwt=4085b42190c19a0a1b41cdca4124fca5
66 http://gdcvault.com/play/1022160/More-Science-Behind-Shaping-Player [16:05 - 18:38]
133 http://boards.na.leagueoflegends.com/en/c/miscellaneous/WV2VqZca-new-champ-select-qa-with-the-dev-team?comment=00350000
146 https://www.reddit.com/r/leagueoflegends/comments/36sgex/soo_anybody_gotten_banned_yet_willing_to_share/crh5e5v
147 https://www.reddit.com/r/leagueoflegends/comments/36sgex/soo_anybody_gotten_banned_yet_willing_to_share/crh5d7s
178 https://ask.fm/RiotLyte/answers/130599286722 ; Page 315
183 https://ask.fm/RiotLyte/answers/130596264898 ; Page 314
262 https://www.reddit.com/r/leagueoflegends/comments/369a4k/reporting_feels_so_useless/crcg867
286 https://www.reddit.com/r/leagueoflegends/comments/369a4k/reporting_feels_so_useless/crccl9s
345 https://www.reddit.com/r/leagueoflegends/comments/369a4k/reporting_feels_so_useless/crc0bro
386 https://ask.fm/RiotLyte/answers/131015276226 ; Page 220
416 https://ask.fm/RiotLyte/answers/130722759618 ; Page 276
418 https://ask.fm/RiotLyte/answers/130722330562 ; Page 277
464 https://ask.fm/RiotLyte/answers/132252998338 ; Page 112
473 https://www.reddit.com/r/leagueoflegends/comments/34rghy/why_does_it_say_the_tribunal_system_has_improved/cqxho0p