Abstract
With the growing demand for personalized marketing, recommender systems have become essential tools to help users quickly discover products or content that match their interests. However, traditional recommendation methods face significant limitations in handling complex user behaviors and sparse data, particularly in accurately capturing relationships among diverse interaction types and higher-order dependencies. To address these challenges, this paper proposes a novel recommendation model based on graph neural networks (MBH-GNN) to optimize personalized marketing strategies. MBH-GNN constructs a multi-behavior interaction graph and employs neighborhood-aware modeling to effectively integrate diverse user-item interaction types (e.g., browsing, favoriting, purchasing), dynamically assigning weights to these behaviors to generate semantically rich embeddings. Furthermore, the model incorporates a high-hop relational learning mechanism to capture long-range user-item dependencies, enhancing its ability to model contextual information. These features enable MBH-GNN to achieve higher recommendation accuracy and diversity in complex scenarios. Experimental results demonstrate that MBH-GNN significantly outperforms existing baseline methods, achieving HR@10 of 0.789 and NDCG@10 of 0.330 on the BeiBei dataset, and HR@10 of 0.773 and NDCG@10 of 0.319 on the Tmall dataset. The model exhibits exceptional robustness and adaptability, particularly in addressing data sparsity and cold-start scenarios. This study offers an efficient and scalable solution for personalized marketing, providing critical theoretical support and practical value for improving recommendation system performance and addressing complex user behavior modeling challenges.