Graph Neural Networks Powered Scientific Paper Recommendation

Scientific paper recommendation systems aim to help researchers discover relevant papers amidst the vast and ever-growing body of literature. With the exponential yearly increase in scientific publications, the demand for effective paper recommendation solutions has become both critical and increasingly challenging. In recent years, deep learning techniques have revolutionized recommender systems, and scientific paper recommendations have naturally integrated these advancements. In this dissertation, we address these challenges through three progressive contributions. First, we enhance traditional content-based methods using Graph Neural Networks (GNNs) by introducing a Graph Convolutional Network-strengthened Topic Modeling (GCN-TM) approach. This method improves upon conventional topic modeling techniques by incorporating community insights through embedding propagation, enriching topic representations, and enabling more robust, community-aware recommendations. Next, we address the dynamic nature of citation networks and their influence on paper embeddings as they evolve over time. To tackle the challenge of recommending relevant papers for works-in-progress, we develop a time-series embedding technique that captures shifts in citation network interactions. This method ensures the recommendation system remains adaptive and contextually relevant, even as the citation landscape changes. Finally, we present a graph learning-based context-aware citation recommendation mechanism that models the decision-making process behind citation selection. To ensure scalability, we construct an overlay citation context network atop the traditional paper-to-paper citation graph. This overlay network acts as a guiding agent, facilitating accurate and context-sensitive citation recommendations during the composition of scientific documents. This dissertation primarily focuses on scientific paper recommendation using graph learning techniques across multiple dimensions, including a hybrid approach to topic modeling for paper recommendation, temporal graph learning, and graph-based context-aware citation recommendation. The proposed methods not only enhance the precision and adaptability of recommendation systems but also provide practical guidance for both research and engineering applications. This work establishes a foundation for future innovations in scientific paper recommendation, equipping researchers with powerful tools to navigate and contribute to the ever-expanding body of scientific knowledge.