Research on the Optimization of Collaborative Filtering Algorithms in Deep Learning-Based E-Commerce Recommendation Systems

With the rapid development of e-commerce, recommendation systems play an important role in enhancing user shopping experiences and increasing corporate profits. However, traditional collaborative filtering algorithms face challenges such as data sparsity, cold start, and high computational complexity, which limit their application in large-scale e-commerce platforms. To address these issues, this paper proposes a deep learning-based collaborative filtering algorithm optimization model, MAGR. This algorithm innovatively optimizes the collaborative filtering approach by leveraging the advantages of deep learning. The model comprehensively enhances the performance and adaptability of the recommendation system through the integration of multimodal features, dynamic user preference capture, and high-order relationship modeling. First, a recommendation framework that integrates multimodal features is constructed, enhancing the representation capabilities of user and product features by incorporating multimodal data such as user behavior sequences, product text descriptions, and image information. Second, an attention mechanism-based embedding learning model is designed to dynamically capture key features of user preferences and eliminate noise interference. Finally, by introducing graph neural networks, complex relational structures are modeled on the user-product interaction graph, further improving the accuracy and generalization ability of recommendations. Experimental results show that the proposed algorithm significantly outperforms traditional collaborative filtering and other deep learning models in recommendation performance on the dataset, addressing data sparsity and cold start issues, and demonstrating strong practical application value.