Graph-Based Retrieval-Augmented Generation (Rag) Systems

Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm for enriching language models with external knowledge. While traditional RAG approaches rely on unstructured text retrieval, there is growing interest in more structured and interpretable retrieval formats. In this context, graph-based methods are gaining traction for their ability to model multi-hop reasoning, semantic relationships, and rich contextual dependencies. These capabilities make them especially valuable for complex generation tasks requiring a deeper understanding and structured knowledge integration.

This article collection aims to explore recent advances in graph-enhanced Retrieval-Augmented Generation (RAG) from a wide range of perspectives. Topics of interest include, but are not limited to, novel Knowledge Graph (KG) construction techniques tailored for generation tasks, retrieval algorithms over structured graphs, reasoning and alignment methods for incorporating graph-based knowledge into generation, and evaluation frameworks for assessing both factuality and coherence. We also welcome application-driven studies in domains such as scientific discovery, healthcare, legal reasoning, and education, where structured knowledge plays a critical role in improving the quality and trustworthiness of generated outputs.

We invite high-quality submissions covering theoretical models, algorithmic innovations, system design, and empirical analysis at the intersection of retrieval, graph learning, and natural language generation. Contributions may span areas such as graph construction and representation learning for RAG, graph-based retrievers, graph-aware generators, and benchmarks for structured generation. This Research Topic provides a dedicated venue for advancing knowledge-grounded generation through structured retrieval, fostering collaboration across the Natural Language Processing (NLP), Information Retrieval (IR), and Graph Learning communities.