Graph Convolution Network and User Interest Modeling for Enhanced Recommendation Systems

In the realm of modern recommendation systems, efficiently handling the vast amount of available data and accurately predicting user preferences remains a significant challenge. Traditional models often struggle with sparse data and fail to capture the dynamic nature of user interests. This paper introduces an innovative solution, the Hybrid User Interest Knowledge Graph and Graph Convolutional Network (HUIKGKCN), which is specifically designed to address these challenges. HUIKGKCN revolutionizes recommendation systems by focusing on the intricacies of user behavior and the complex web of user-item relationships. The model employs Long Short-Term Memory (LSTM) networks and advanced attention mechanisms to delve into the user's historical interactions, extracting refined interest features that reflect both the temporal dynamics and the varying importance of each action. This nuanced approach ensures a more accurate and personalized user profile. Moreover, the model overcomes the limitations of traditional embedding methods like TransR by incorporating the GC-OTE embedding method. This novel approach adeptly handles various complex relational scenarios, including many-to-one, one-to-many, and many-to-many relationships, thus significantly enhancing the model's ability to understand and predict user preferences. The unique strength of HUIKGKCN lies in its use of heterogeneous propagation. By integrating knowledge graphs with user-item interaction information, the model not only reduces knowledge noise but also encodes collaborative signals more effectively. The use of a knowledge-aware attention function further refines the process, focusing on essential information and minimizing irrelevant data. The efficacy of HUIKGKCN has been rigorously tested and validated on two diverse datasets, Book_Crossing and Last.FM. The results are impressive, showing a marked improvement over several baseline models across various critical metrics. These outcomes highlight the model's robustness and its ability to adapt to different types of data, making it a highly versatile tool in the ever-evolving field of recommendation systems.