Graph-Based Re-ranking: Emerging Techniques, Limitations, and Opportunities
Md Shahir Zaoad, Niamat Zawad, Priyanka Ranade, Richard Krogman, Latifur Khan, James Holt
TL;DR
The paper surveys graph-based re-ranking approaches for Retrieval Augmented Generation, focusing on how Graph Neural Networks leverage relational knowledge to refine initial retrievals. It presents a taxonomy of models across Pointwise, Pairwise, and Listwise paradigms, and surveys graph-construction strategies ranging from document- to entity-level graphs and knowledge-graph augmentations. Key contributions include a synthesis of architectures (e.g., GAR, KGPR, GraphMonoT5, AMR-based G-RAG) and a discussion of external knowledge integration, subgraph construction, and cross-encoder combinations. The work highlights the lack of standardized benchmarks and reproducibility as major barriers and provides recommendations to advance the field with static benchmarks, consistent evaluation, and robust graph-generation pipelines.
Abstract
Knowledge graphs have emerged to be promising datastore candidates for context augmentation during Retrieval Augmented Generation (RAG). As a result, techniques in graph representation learning have been simultaneously explored alongside principal neural information retrieval approaches, such as two-phased retrieval, also known as re-ranking. While Graph Neural Networks (GNNs) have been proposed to demonstrate proficiency in graph learning for re-ranking, there are ongoing limitations in modeling and evaluating input graph structures for training and evaluation for passage and document ranking tasks. In this survey, we review emerging GNN-based ranking model architectures along with their corresponding graph representation construction methodologies. We conclude by providing recommendations on future research based on community-wide challenges and opportunities.