Search papers, labs, and topics across Lattice.
This paper introduces Low-Rank Graph Message Prompting (LR-GMP), a novel approach to graph prompt learning that unifies prompting across all graph components (nodes, edges, weights). By reinterpreting existing Graph Data Prompts (GDPs) through a Graph Message Prompt (GMP) paradigm, LR-GMP achieves a compact and effective prompt representation using low-rank matrices. Experiments on benchmark datasets demonstrate that LR-GMP outperforms existing GDPs in generalization and robustness due to its unified prompting strategy.
Forget hand-crafting prompts for each graph component – LR-GMP learns a single, low-rank prompt that works across nodes, edges, and weights, boosting generalization.
Graph Data Prompt (GDP), which introduces specific prompts in graph data for efficiently adapting pre-trained GNNs, has become a mainstream approach to graph fine-tuning learning problem. However, existing GDPs have been respectively designed for distinct graph component (e.g., node features, edge features, edge weights) and thus operate within limited prompt spaces for graph data. To the best of our knowledge, it still lacks a unified prompter suitable for targeting all graph components simultaneously. To address this challenge, in this paper, we first propose to reinterpret a wide range of existing GDPs from an aspect of Graph Message Prompt (GMP) paradigm. Based on GMP, we then introduce a novel graph prompt learning approach, termed Low-Rank GMP (LR-GMP), which leverages low-rank prompt representation to achieve an effective and compact graph prompt learning. Unlike traditional GDPs that target distinct graph components separately, LR-GMP concurrently performs prompting on all graph components in a unified manner, thereby achieving significantly superior generalization and robustness on diverse downstream tasks. Extensive experiments on several graph benchmark datasets demonstrate the effectiveness and advantages of our proposed LR-GMP.
