Classical approaches to finding related chess positions are complex to implement and not very efficient when applied to large game databases. Moreover, they only find completely matching positions, which excludes others that may be more similar in content but do not match exactly.
In this thesis, we have developed an easy-to-use tool that enables the identification of related chess positions based on a given position. The information extraction methods used allow for fast and efficient searching within large databases of games containing numerous positions. The tool considers both static and dynamic similarity in its searches. We have focused on maximising the identification of positions in the opening and early part of the middlegame.
We have found that some of the used features, such as pawn structures and center types, play a key role in identifying related positions. The program has been successfully tested on a large database. In an experiment in which the participants were chess experts, we found a high agreement between the similarity ratings of the chess players and the similarity ratings of the program. The results of the experiment confirmed the system's correct functioning and its ability to efficiently identify related positions.
|
