DeepView: Computational Tools for Chess Spectatorship by Greg Borenstein, Prof. Kevin Slavin, Grandmaster Maurice Ashley.
From the post:
DeepView is a suite of computational and statistical tools meant to help novice viewers understand the drama of a high-level chess match through storytelling. Good drama includes characters and situations. We worked with GM Ashley to indentify the elements of individual player’s styles and the components of an ongoing match that computation could analyze to help bring chess to life. We gathered an archive of more than 750,000 games from chessgames.com including extensive collections of games played by each of the grandmasters in the tournament. We then used the Stockfish open source chess engine to analyze the details of each move within these games. We combined these results into a comprehensive statistical analysis that provided us with meaningful and compelling information to pass on to viewers and to provide to chess commentators to aid in their work.
The questions we answered include:
- How do you describe different players’ style of chess?
- Can you describe the matchup between two players in a way that sets expectations for a match?
- Are some players more likely to win short or long games?
- Are some players particularly good or bad in the late phase of the game?
- How likely is a player with a higher ranking to actually beat a lower player?
- Can we reduce a chess position to a single meaningful score that would be easily understood?
- Can we detect the most interesting games going on in a tournament with hundreds of players?
In addition to making chess more accessible to novice viewers, we believe that providing access to these kinds of statistics will change how expert players play chess, allowing them to prepare differently for specific opponents and to detect limitations or quirks in their own play.
Further, we believe that the techniques used here could be applied to other sports and games as well. Specifically we wonder why traditional sports broadcasting doesn’t use measures of significance to filter or interpret the statistics they show to their viewers. For example, is a batter’s RBI count actually informative without knowing whether it is typical or extraordinary compared to other players? And when it comes to eSports with their exploding viewer population, this approach points to rich possibilities improving the spectator experience and translating complex gameplay so it is more legible for novice fans.
A deeply intriguing notion of mining data to extract patterns that are fashioned into a narrative by an expert.
Participants in the games were not called upon to make explicit the tacit knowledge they unconsciously rely upon to make decisions. Instead, decisions (moves) were collated into patterns and an expert recognized those patterns to make the tacit knowledge explicit.
Outside of games would this be a viable tactic for knowledge retention? Not asking employees/experts but recording their decisions and mining those for later annotation?