This study presents a convolutional neural network-based evaluation method for Internet course teaching modes, aiming to optimize the teaching approach and enhance its effectiveness. The initial step involves determining the evaluation criteria for each subject and constructing a comprehensive evaluation index system for Internet course teaching modes. Teaching resources are then decomposed using mutual information and contrast angle principles, extracting essential features of Internet course teaching materials. These features are fed into a decision tree classifier to categorize the teaching resources. Leveraging convolutional neural networks with enhanced learning parameters, the evaluation scores for the Internet course are derived, ultimately establishing the teaching model's evaluation level. Experimental findings demonstrate the efficacy of this evaluation model. In the test dataset, the evaluation model's decision coefficient is 0.97, indicating a robust fitting effect. In the training dataset, the training error remains low at 0.26, and the evaluation accuracy exceeds 98.61%. The evaluated Internet course showcases commendable teaching outcomes.