Integrating Large Language Models and Graphs for Content Recommendation
Open Access DepositedLarge Language Models (LLMs) powered by deep learning have dramatically advanced natural language processing, enabling machines to comprehend and generate human language with exceptional precision. Simultaneously, graph-based methodologies, such as graph neural networks (GNNs) and knowledge graphs, have demonstrated substantial potential in capturing and reasoning over complex data structures. This dissertation explores the synergy between these two approaches, highlighting their complementary strengths. While LLMs excel in interpreting unstructured text, graph-based techniques are proficient in representing structured relationships. In fact, recent studies have shown that the Transformer architecture underlying LLMs can itself be interpreted as a form of graph neural network, where self-attention acts as message passing on a fully connected graph. This perspective situates LLMs and GNNs within a unified theoretical framework, revealing that they are not just complementary but intrinsically related. By integrating knowledge graphs and large language models, this work focuses on key challenges in recommendation problems. The dissertation is organized as a sequence of research steps
first, advancing the semantic and reasoning capabilities of language models, and next, examining how these improvements can be applied to ranking formulations. Finally, demonstrating their impact on recommendations. This staged progression highlights how insights from LLMs and graph-based methods accumulate over multiple studies, collectively enhancing recommendation performance.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.