Choosing the Right Path: A Comparative Study of CNN, RNN, and Transformer Models for Sequential Recommendation Systems

As the Internet continues to evolve rapidly, the accuracy and efficiency of recommendation systems are crucial for Internet platforms. Concurrently, the continuous advancement of deep learning has led to the emergence of various recommendation system models based on Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and attention mechanisms. This model diversity presents a challenge for platforms in selecting the appropriate model. This paper comprehensively analyses the key modules within models based on these three architectures. Furthermore, representative models from each architecture are evaluated on a benchmark dataset, comparing both business performance metrics and computational resource consumption. The goal is to guide platforms in choosing the most suitable model based on their specific requirements rather than relying solely on the novelty of the architecture or single performance metrics.