A User Purchase Motivation-Aware Product Recommender System

• 成果类型: Article; Early Access
• 署名作者: Xu, Jiarong; Wang, Jiaan; Zhang, Hongzhe; Lu, Tian
• 署名单位: Fudan University; The Chinese University of Hong Kong, Shenzhen; The Chinese University of Hong Kong, Shenzhen; Arizona State University; Arizona State University-Tempe
• 刊物名称: INFORMATION SYSTEMS RESEARCH
• ISSN/ISSBN: 1047-7047; 1526-5536
• DOI: 10.1287/isre.2024.1028
• 发表日期: 2026-02-06
• 关键词: recommendation; user purchase motivation; graph neural network; Deep learning; management; STABILITY; variety; signals; CHOICE; MODEL

摘要

Product recommender systems play a critical role in predicting users' future interests based on their historical behavior, but meeting users' all-time needs requires a deep understanding of their essential purchase motivations. However, this realm is dominated by methods that do not explicitly specify motivation categories or suffer from low data efficiency due to heavy reliance on extensive auxiliary data. To address this gap, we introduce a comprehensive framework for categorizing product purchase motivations. Building on this framework, we identify two key purchase motivations rooted in users' inherent interests that drive purchasing decisions: stable preference and exploratory intent. Methodologically, we introduce a measure STB to explicitly identify which of the two motivations drives the purchase of a product item, relying solely on historical behavior sequences and items' intrinsic attributes. Leveraging this measure, we propose User PurchaSe moTivation-Aware Recommendation, a novel product recommendation method that establishes a clear inference chain for these two motivations, thereby enhancing overall recommendation performance. Extensive experiments on three real-world e-commerce recommendation scenarios demonstrate our model's superiority over stateof-the-art recommendation benchmarks. Further empirical analyses offer intriguing insights and highlight the substantial progress our method achieves in tackling the challenging task of recommending items driven by exploratory intent.