When looking for a best move in a given position, a chess player explores in his mind a tree of possible continuations of the game. To cope with a large combinatorial complexity of this tree, the player uses typical chess motifs, such as double attacks or pinned pieces. In this thesis we attempt to automatically detect from the player's eye movement the motifs that the player is using during problem solving. We developed a formula that converts eye tracking data obtained from problem solving, into a degree of membership for predefined chess motifs in the position. Results were analysed and compared with retrospections of chess players, which were obtained immediately after the problem solving experiment. Then the time series of motifs were adjusted in different ways, so they are more convenient to use with machine learning algorithms. We trained a neural network to predict players’ chess moves from their eye movements. The developed method for motif detection seems to work promising, however it has a disadvantage of not being able to perform in positions where very similar motifs exist.
|
