Temporal attention graph-optimized networks for sequential recommendation

• Main Author: Pathak, Siddhant
• Other Authors: Ke Yiping, Kelly
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Graph neural networks; Recommendation systems; Sequential recommendation
• Online Access: https://hdl.handle.net/10356/175250
• Institution: Nanyang Technological University

Sequential recommendation systems are pivotal in enhancing user experience by providing personalized suggestions based on historical data. However, traditional models often disregard the temporal dynamics of user-item interactions, which can lead to static and outdated recommendations that fail to reflect current user preferences. To address this critical gap, this study introduces a novel architecture known as Temporal Attention Graph-Optimized Networks (TAGON), which integrates Graph Neural Networks (GNNs) with an innovative temporal attention mechanism. TAGON is designed to dynamically adapt to both the sequential and temporal aspects of user interactions, thereby offering more accurate and relevant recommendations. This research extends the capabilities of GNNs by incorporating a temporal dimension, allowing the system to track and predict the evolving preferences of users over time. The effectiveness of TAGON is rigorously tested against established benchmarks on several real-world datasets, demonstrating superior performance in terms of precision, recall, and user satisfaction. The results indicate that incorporating temporal dynamics into the recommendation process significantly improves the relevance and timeliness of the suggestions offered to users. This study not only validates the importance of temporal considerations in modeling user- item interactions but also sets a new standard for the design of adaptive recommendation systems in continuously evolving digital environments. Overall, TAGON represents a significant advancement in the field of recommendation systems, promising enhanced user engagement through more intelligent and context-aware recommendations.