Due to the challenges associated with mastering fundamental piano playing techniques, the inefficiency of self-guided learning, and the prohibitive cost of one-on-one instruction, many novices abandon their musical pursuits prematurely. Our research addresses these issues by enhancing music feature extraction methods through artificial intelligence modeling and developing a piano-playing ability evaluation system. This system leverages an attention mechanism and an LSTM neural network model to assess a player’s abilities based on rhythm, thematic prominence, and musical expression within various levels of piano scores. By analyzing sample tracks from the Thompson Simple Piano Tutorial, our system demonstrates robust performance, achieving an overall F-Measure above 0.9 with an average value of 0.9641. These results indicate that the evaluation system offers precise assessments and can significantly aid piano instruction, providing learners with reliable feedback on their progress.