The School of Computer Science is pleased to present…
Abstract:
Graphs are a widely used way to represent data in a structured way, in which entities are connected through relationships. They have become quite essential in several domains, such as social networks, collaboration networks, knowledge base networks, etc., where link prediction and node classification are key learning tasks. At the same time, recent advancements in Large Language Models (LLMs) suggest that, when appropriately represented, LLMs can reason over structured information, improving overall effectiveness.
This study presents two novel frameworks leveraging LLMs in graph learning tasks. The first work introduces an LLM-Guided Framework for Link Prediction in Homogeneous Graphs on purely structural information. Whereas the second work introduces KNodeLLM: Knowledge Graph Node Classification using LLMs, investigating semantic reasoning over Knowledge Graphs. Experimental results across multiple benchmark datasets show that the proposed approaches achieve significant improvements over traditional methods or, at least, are on par with state-of-the-art results, with an improvement of up to 15% in node classification, specifically, and strong performance in link prediction with minimal supervision.
Keywords: Link Prediction, Node Classification, Large Language Models, Homogeneous Graphs, Knowledge Graphs
