- 學位論文
基於機器學習方法的中國象棋棋譜異常檢測
Anomaly Detection of Game Records in Chinese Chess based on Machine Learning Method
摘要
大量且品質穩定的資料有助於建立象棋人工智慧模型，然而網路象棋對戰平台上的棋譜有許多不合理的走步以及錯誤的勝負結果，所以本研究想要透過資料前處理的方式排除或修正異常資料，從而建立更加可靠的象棋人工智慧模型。 本研究首先將棋局分割成開局、中局、殘局三個部分。開局的部分會利用有效步數及審局函數來建立機器學習模型，分類出有不合理走步的開局。殘局的部分是通過兵種組合及審局函數來分析，並排除或修正原始資料上的異常勝負和。至於中局是本研究尚未解決的部分，也是未來可以研究的方向。透過本論文的方法，能夠有效提升棋譜資料的品質，並能提高後續使用此資料的人工智慧模型準確率。
並列摘要
A large amount of high-quality data is essential for building reliable AI models for chess. However, many online chess platform game records contain unreasonable moves and incorrect game outcomes. Therefore, this study aims to eliminate or correct abnormal data through data preprocessing to establish a more reliable AI model for chess. This study first divides the chess games into three phases: the opening, middlegame, and endgame. In the opening phase, a machine learning model is constructed using valid moves and evaluation functions to classify and identify unreasonable moves. For the endgame phase, piece combinations and evaluation functions are used to analyze and exclude or correct abnormal results in the original data. The middlegame phase remains unresolved in this study and represents a potential area for future research. The methods presented in this thesis effectively improve the quality of game record data and enhance the accuracy of AI models that utilize this data.