The School of Computer Science is pleased to present…
An LLM-Guided Framework for Link Prediction in Homogeneous Graphs
MSc Thesis Proposal by: Atul Kumar
Date: Tuesday, July 22, 2025
Time: 2:30 PM
Location: Essex Hall, Room 122
As social networks continue to grow rapidly, link prediction has become vital in network analysis, estimating the likelihood of connections between unconnected nodes based on structural similarity. This study explores the intersection of Large Language Models (LLMs) and Graph Learning, with a particular focus on Link Prediction tasks on Homogeneous Networks. While numerous studies have leveraged LLMs for Knowledge Graphs, Heterogeneous Graphs, and Text-Attributed Graphs, the use of LLMs for Homogeneous Graphs with no textual information, to the best of our knowledge, is still an understudied area, which this research aims to explore. Our proposed framework explores the use of topological features of the graph, showing that LLMs can effectively capture and reason over connectivity patterns without relying on any node or edge text, which fundamentally frees LLM-based Link Prediction from the constraints of "graph-with-text."
We leverage prompting strategies, including Zero-Shot, One-Shot, and Few-Shot learning, which allow LLMs to perform the task when the available data is scarce. Unlike traditional supervised models like VERSE, GCN, and SEAL that require full retraining for new graphs, our prompting-based framework generalizes from minimal examples across different network types. Additionally, we fine-tune LLMs on domain-specific networks, which further improve performance and enhance generalization. Experimental results across multiple datasets demonstrate that our approach achieves on par with state-of-the-art results for the Facebook dataset, while similarly achieving strong outcomes across other datasets.
Internal Reader: Dr. Xiaobu Yuan
External Reader: Dr. Bharat Maheshwari
Advisor: Dr. Asish Mukhopadhyay
Co-Advisor: Dr. Zamilur Rahman