REVIEW 15 cited by
Talk like a Graph: Encoding Graphs for Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Graphs are a powerful tool for representing and analyzing complex relationships in real-world applications such as social networks, recommender systems, and computational finance. Reasoning on graphs is essential for drawing inferences about the relationships between entities in a complex system, and to identify hidden patterns and trends. Despite the remarkable progress in automated reasoning with natural text, reasoning on graphs with large language models (LLMs) remains an understudied problem. In this work, we perform the first comprehensive study of encoding graph-structured data as text for consumption by LLMs. We show that LLM performance on graph reasoning tasks varies on three fundamental levels: (1) the graph encoding method, (2) the nature of the graph task itself, and (3) interestingly, the very structure of the graph considered. These novel results provide valuable insight on strategies for encoding graphs as text. Using these insights we illustrate how the correct choice of encoders can boost performance on graph reasoning tasks inside LLMs by 4.8% to 61.8%, depending on the task.
Forward citations
Cited by 15 Pith papers
-
C-RE-ACT: Causal RE-ACTing Agent for O-RAN Forensic Triage
An agentic O-RAN triage system that ranks root causes via SAM causal discovery and graph soft-prompting claims 89% top-3 accuracy on 140 testbed experiments.
-
Lost in Serialization: Invariance and Generalization of LLM Graph Reasoners
Fine-tuned graph reasoners become more stable to node relabeling but more brittle to edge ordering and format, and they do not consistently beat base models on new spectral graph tasks.
-
Modeling Code: Is Text All You Need?
A GNN-encoded LLVM IR graph, prepended as soft prompts to a frozen code LLM, improves accuracy on device mapping, algorithm classification, vulnerability detection, and code translation tasks.
-
Enhancing LLM Agent Safety via Causal Influence Prompting
CIP, which makes LLM agents construct and refine a causal influence diagram before acting, raises refusal rates on harmful tasks in three agent-safety benchmarks.
-
Are Large Language Models Good Temporal Graph Learners?
TGTalker prompts LLMs with the most recent edges and neighbor history of a temporal graph, achieving link prediction accuracy competitive with specialized temporal graph neural networks on five real-world datasets.
-
G1: Teaching LLMs to Reason on Graphs with Reinforcement Learning
Reinforcement learning on synthetic graph-theoretic tasks sharply improves LLM graph reasoning, transferring to larger graphs, new encodings, and real-world tasks.
-
HierPromptLM: A Pure PLM-based Framework for Representation Learning on Heterogeneous Text-rich Networks
HierPromptLM uses hierarchical text prompts and two language-model pretraining tasks to jointly encode node text and heterogeneous graph structure, outperforming prior text-plus-GNN baselines on DBLP and OAG.
-
Bridging Visualization and Optimization: Multimodal Large Language Models on Graph-Structured Combinatorial Optimization
Feeding graph images to GPT-4o and refining its node choices with local search outperforms centrality and GNN baselines on influence maximization and network dismantling.
-
Decomposing Visual Classification: Assessing Tree-Based Reasoning in VLMs
Tree-based reasoning prompts consistently underperform standard zero-shot prompting for VLM image classification on GTSRB and CIFAR-10 across three models.
-
DGP: A Dual-Granularity Prompting Framework for Fraud Detection with Graph-Enhanced LLMs
DGP condenses neighbors into coarse-grained summaries while preserving the target node's full text, improving LLM-based fraud detection by up to 6.8 absolute AUPRC points over state-of-the-art methods.
-
Graph Repairs with Large Language Models: An Empirical Study
Open-source LLMs can format graph repairs well and delete the violating edge, but exact-matching the intended repair is rare (up to 38%), and the validity metric used is trivially satisfied by deleting any edge.
-
Network-informed Prompt Engineering against Organized Astroturf Campaigns under Extreme Class Imbalance
Frozen LLMs with balanced retrieval-augmented prompting detect astroturf campaigns better than GNN baselines on a 2016 US election dataset, but the reported margins are overstated.
-
REBot: From RAG to CatRAG with Semantic Enrichment and Graph Routing
A category-routed hybrid of RAG and knowledge-graph retrieval answers Vietnamese university-regulation questions with F1 98.89% on the authors' own dataset — about 0.2 points above plain RAG.
-
DistRAG: Towards Distance-Based Spatial Reasoning in LLMs
Retrieving distance facts from a spatial graph improves LLM answers to direct and nearest-city distance questions, while complex distance-comparison questions remain unsolved.
-
LLMs Between the Nodes: Community Discovery Beyond Vectors
CommLLM, a two-step graph-to-text plus LLM prompting method, reports high NMI on six small networks, but its evaluation omits standard community-detection baselines and relies on a prompt tuned on one test set.
Discussion (0). Continue with ORCID to comment.