Complexity Science Hub and Centro Ricerche Enrico Fermi (CREF) researchers examined the similarities between various chess openings using actual data from an online chess platform. They created a new classification technique to support the existing classification based on these similarities.
Giordano De Marzo of the Complexity Science Hub and the Centro Ricerche Enrico Fermi (CREF) explains, “To find out how similar chess openings actually are to each other—meaning in real game behavior—we drew on the wisdom of the crowd. In order to determine who plays which opening games, the researchers examined 3,746,135 chess games, 18,253 players, and 988 different openings from the chess platform Lichess. It makes sense that they would be comparable if a number of players consistently chose two similar opening games. Opening games that are so well-liked that they frequently occur alongside most others were disqualified.
“We also limited the players we examined to those with ratings on the Lichess platform that were higher than 2,000. Vito D.P. explains that complete beginners could arbitrarily play any opening game, which would skew our analyses. For the Complexity Science Hub, contact Servedio. The study “Quantifying the complexity and similarity of chess openings using online chess community data” has been published in Scientific Reports.
“Because of how something has grown over time, it contains a lot of relevant information. Our clustering reflects a new order that is comparable to the existing one and can supplement it by telling players how similar openings are to each other.”
Vito D.P. Servedio of the Complexity Science Hub.
Ten distinct clusters were identified.
The researchers discovered that particular opening games tend to cluster in this manner. Actual similarities in playing behavior caused ten different clusters to stand out. According to De Marzo, “And these clusters don’t necessarily correspond with the standard classification of chess openings.”.
For instance, the same players frequently played specific opening games from various classes. These tactics must therefore share some characteristics, even though they are categorized in different classes. Therefore, they are all part of the same cluster. Each cluster, then, stands for a specific playing style, such as being very offensive or very defensive. The classification method that the researchers have created here can also be used to classify other games like Go or Stratego that are similar to chess.
Amplify the conventional classification.
Chess openings typically consist of fewer than 20 moves. An opening can be described as open, half-open, closed, or irregular, depending on which pieces are moved first. The ECO Code, also known as the Encyclopedia of Chess Openings, classifies them into five major categories: A, B, C, D, and E.
This contains a lot of helpful information because it has historically evolved. By demonstrating to players how similar openings actually are to one another, our clustering represents a new order that is close to the one currently in use. After all, something that has a history of growth cannot be completely rearranged. “You cannot claim that A20 has now become B3. De Marzo continues, “That would be like trying to communicate in another language.
Players and opening games are rated.
The researchers’ approach also enabled them to gauge a player’s skill level and the difficulty of a specific opening game. The fundamental premise is that an opening game will probably be simple if it is played by a lot of players. Therefore, they looked at which opening games were played the most and by whom. This provided the researchers with a gauge for the complexity (=difficulty) and fitness (=skill) of an opening game. There was a strong correlation between these and the players’ ratings on the chess platform.
“On the one hand, this highlights the significance of our two newly introduced measures but also the accuracy of our analysis,” says Servedio. The researchers consulted with a renowned chess grandmaster who prefers to remain anonymous in order to guarantee the relevance and validity of these results from a chess theory perspective.
More information: Giordano De Marzo et al, Quantifying the complexity and similarity of chess openings using online chess community data, Scientific Reports (2023). DOI: 10.1038/s41598-023-31658-w
