Personalized Recommendation System for E-Commerce Websites Based on Multi-Scale Graph Neural Network

With the prosperity of the e-commerce industry, personalized recommendation systems are becoming increasingly important. However, traditional recommendation algorithms are limited in handling complex user behavior and dynamic interactions. Therefore, this paper innovatively proposes a recommendation framework based on multi-scale graph neural networks to accurately capture multi-level information in user-product interaction. This framework constructs a complex graph structure that covers users, products and diverse relationships. It utilizes hierarchical attention networks and residual linear graph convolutional networks to aggregate features from local to global through multi-scale graph convolution, and embeds emotional information to deeply analyze user preferences, product characteristics, and their complex relationships. At the same time, an attention mechanism is introduced to dynamically adjust the weight of information, enhancing the personalization and accuracy of recommendations. The experimental results show that compared to traditional recommendation algorithms and single-scale graph neural network models, the multi-scale graph neural network recommendation system proposed in this paper can significantly improve the accuracy, diversity and user satisfaction of recommendation results. This research not only enriches the application of graph neural networks in the field of recommendation systems, but also provides more intelligent and efficient personalized recommendation solutions for e-commerce websites.