The game of Go, renowned for its strategic depth, has been a central focus in both competitive gaming and artificial intelligence (AI) research. This paper explores the application of deep learning techniques to recognize the playstyles of human Go players, a task that offers valuable insights into complex human decision-making. The study employs a neural network architecture, leveraging convolutional layers, residual connections, and attention mechanisms, to categorize player styles into three specific groups: Fighting, Balanced, and Territorial. Trained on a dataset of 70,000 original Go game records, which was expanded through data augmentation to 483,712 samples, the model achieved a high testing accuracy of 82.6 % on a separate, unseen dataset of 10,000 game records. These results demonstrate the model’s effectiveness in accurately distinguishing between these playstyles and its strong generalization capability, with a final validation accuracy of 81.03 %. The model successfully identifies players preferred playing styles, revealing consistent preferences aligned with Go literature. This work contributes to the field of behavioral stylometry by showcasing how deep learning can be applied to complex strategic behaviors, with potential implications for player modeling and AI-human collaboration in various strategic contexts.
扫码关注我们
求助内容:
应助结果提醒方式:
