Trustworthy GNNs with LLMs: A Systematic Review and Taxonomy

TL;DR

Trustworthiness of GNNs is critical in sensitive domains, and this paper investigates how integrating LLMs can improve reliability, robustness, privacy, and reasoning. It proposes a four-dimensional taxonomy and systematically surveys representative LLM-GNN methods under these categories, highlighting how LLMs can augment graph structure, node semantics, and explainability. The review also discusses privacy threats, robustness under adversarial perturbations, and the current limitations of privacy-preserving approaches in LLM-GNNs. It concludes with future directions, including low-text-dependency and multimodal LLM-GNNs, fairness considerations, and opportunities from rapidly evolving LLMs.

Abstract

With the extensive application of Graph Neural Networks (GNNs) across various domains, their trustworthiness has emerged as a focal point of research. Some existing studies have shown that the integration of large language models (LLMs) can improve the semantic understanding and generation capabilities of GNNs, which in turn improves the trustworthiness of GNNs from various aspects. Our review introduces a taxonomy that offers researchers a clear framework for comprehending the principles and applications of different methods and helps clarify the connections and differences among various approaches. Then we systematically survey representative approaches along the four categories of our taxonomy. Through our taxonomy, researchers can understand the applicable scenarios, potential advantages, and limitations of each approach for the the trusted integration of GNNs with LLMs. Finally, we present some promising directions of work and future trends for the integration of LLMs and GNNs to improve model trustworthiness.

Figures (7)

• Figure 1: Applications of the integration of graphs and LLMs have driven the increased demand for model trustworthiness.
• Figure 2: A taxonomy of trustworthy GNNs with LLM.
• Figure 3: Train an LLM-based edge predictor to compute candidate edges; use a prompting approach to allow the LLM to evaluate the augmented adjacency matrix and determine the final edges.
• Figure 4: The illustration of LLMs integrating GNNs for improved robustness: (a) LLM4RGNN method for robustness enhancement; (b) LLMGRobustness is evaluated from two perspectives.
• Figure 5: Knowledge Distillation Framework from LLM to Graph Model: LLMs generate pseudo labels for the LM model while simultaneously guiding the graph in identifying key nodes and key links through a supervised approach. Finally, knowledge distillation is employed to help the student model achieve superior performance