Topology of Reasoning: Retrieved Cell Complex-Augmented Generation for Textual Graph Question Answering
Sen Zhao, Lincheng Zhou, Yue Chen, Ding Zou

TL;DR
This paper introduces TopoRAG, a novel framework that enhances reasoning over textual graphs by modeling multi-dimensional topological structures, leading to improved question answering performance.
Contribution
The work proposes lifting textual graphs into cellular complexes and developing a topology-aware retrieval and reasoning mechanism, capturing higher-dimensional dependencies for better inference.
Findings
Outperforms existing methods on textual graph question answering tasks.
Effectively models cycles and relational loops in graphs.
Improves reasoning accuracy through topological context.
Abstract
Retrieval-Augmented Generation (RAG) enhances the reasoning ability of Large Language Models (LLMs) by dynamically integrating external knowledge, thereby mitigating hallucinations and strengthening contextual grounding for structured data such as graphs. Nevertheless, most existing RAG variants for textual graphs concentrate on low-dimensional structures -- treating nodes as entities (0-dimensional) and edges or paths as pairwise or sequential relations (1-dimensional), but overlook cycles, which are crucial for reasoning over relational loops. Such cycles often arise in questions requiring closed-loop inference about similar objects or relative positions. This limitation often results in incomplete contextual grounding and restricted reasoning capability. In this work, we propose Topology-enhanced Retrieval-Augmented Generation (TopoRAG), a novel framework for textual graph question…
Peer Reviews
Decision·ICLR 2026 Poster
Introducing cell complex topology into RAG for textual graphs is new. Prior GraphRAG methods (e.g., G- Retriever, GNN-RAG, SubgraphRAG) restrict reasoning to pairwise edges, while TopoRAG models cyclic and higher-order relations. For a formal introduction. the paper carefully defines cell complexes, homology, and the lifting procedure, demonstrating a strong understanding of topological deep learning foundations. Besides, the whole pipline proposed in the paper provides a systematic way to injec
While using cell complexes for retrieval is novel, the reasoning stage (message passing with faces/cofaces) resembles existing simplicial or cell complex GNNs (e.g., CWN, SAN, CellNN). It would be better to integrate clarification about how its update scheme differs algorithmically from those prior works. Building and storing cell complexes (especially identifying all 2-cells from cycles) may be expensive for large textual graphs. While topology sounds interpretable, the paper lacks visualizatio
1. The paper is the first to introduce cellular complexes into the GraphRAG framework, explicitly modeling 2-cells to capture closed-loop topological dependencies that are ignored by traditional node-and-edge paradigms. 2. The method design is logically clear, with a coherent pipeline progressing from cellular lifting to topology-aware retrieval, multi-dimensional reasoning, and generation.
1. All fundamental cycles—regardless of semantic relevance—are lifted into 2-cells, which may introduce noise or spurious dependencies due to meaningless loops. 2. The experiments use datasets that involve multi-hop reasoning, but their graph structures inherently contain many cyclic dependencies. The paper does not evaluate on purely chain-like or acyclic reasoning tasks, making it unclear whether TopoRAG remains advantageous or introduces redundancy in non-cyclic scenarios. 3. The impact of
- The proposed retrieval method enables retrieval at different granular levels, which can capture higher-dimensional topological and relational dependencies.
- **Comparing Methods are Not SOTA Models.** The authors compare with models like G-Retriever, SubgraphRAG, and GraphToken, but these are not state-of-the-art (SOTA). Some related work with better performance, such as DoG [1] and GCR [2], is not cited or compared. These models achieved better results on WebQSP than the proposed TopoRAG. - **Selected Datasets and Their Suitability.** The selected datasets (ExplaGraphs and SceneGraph) may not adequately demonstrate TopoRAG's ability to capture hi
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Graph Neural Networks · Topic Modeling · Multimodal Machine Learning Applications
