.png){kind=icon}
Return
Towards context-aware graph representation learning: Adaptive node aggregation with LLMs
.png){kind=icon}
PRE
AI
DOI：10.1016/j.artint.2026.104584.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Representation learning on graph-structured data is a crucial research area in graph machine learning, with widespread application in real-world domains such as social networks and recommender systems. Traditional graph neural networks (GNNs) typically rely on static graph aggregation functions (e.g., mean, sum) applied over shallow textual features, which limits their ability to model complex semantic interactions among nodes. Recent advances have introduced large language models (LLMs) to enhance node features through deep semantic encoding. However, most existing work treats LLMs as static encoders, overlooking their potential to reason about inter-node relationships. In this work, we propose LLM-based Tailored Node Aggregation (LTNA), a novel approach that leverages the in-context reasoning capabilities of LLMs to generate adaptive, semantics-driven aggregation strategies for each node. Rather than relying on a fixed rule or global function, LTNA prompts the LLM to infer the relational meaning between a node and its neighbors (e.g., supportive, critical, background) and outputs a customed aggregation strategy based on the inferred semantics. This enables message passing in GNNs to move beyond numeric aggregation and become a context-sensitive semantic operation. By jointly leveraging node textual attributes and graph structure, LTNA enhances the expressiveness of node representations and improves performance across downstream tasks. Extensive experiments on five benchmark datasets demonstrate that LTNA consistently outperforms state-of-the-art GNN models, achieving an average improvement of 7.95% in classification accuracy and 9.90% in Micro-F1 score.
Keywords:
Text-attributed graph
Graph neural network
Large language model
Context-aware aggregation
Journal
A
IF:
4.6
Papers:
103
Citations:
1
Organization
View more organizations
Cited Papers
No cited papers available