A Graph Bottleneck and Masked Feature Interaction Framework for Multi-Objective Optimization in E-commerce Recommendation Systems

This study aims to enhance the accuracy of multi-objective prediction in e-commerce recommendation systems, specifically focusing on user behaviors such as clicks, cart additions, and purchases.Utilizing the OTTO dataset, we developed a recommendation system that integrates advanced feature engineering, the ESM2 framework for multi-task learning, and MaskNet for capturing high-order feature interactions.To further improve prediction accuracy, a powerweighted fusion mechanism was implemented, combining clickthrough rate and conversion rate predictions.The proposed model demonstrates superior performance over traditional approaches, offering a more effective solution for handling complex user behaviors in e-commerce environments.