TPGC: Dual-Prior Prompt Initialization for Multi-Task Graph Pre-Training Released
Summary:TPGC, a novel graph prompt learning framework, introduces a dual-prior prompt initialization solution that explicitly models the synergy between task priors and structural priors. This framework includes a Task-Prior Injection Module and a Structure-Prior Injection Module, which extract task optimization preferences and transferable global structural context from an auxiliary graph, respectively. Experiments on six mainstream benchmarks demonstrate that TPGC achieves superior performance in node
Background and Challenges
Graph Prompt Learning is an effective paradigm for adapting pre-trained graph models to downstream tasks. However, existing multi-task graph pre-training frameworks typically use randomly initialized prompts, leading to poor alignment between the prompt space, pretext objectives, and graph structural characteristics. This significantly weakens the task relevance, structural awareness, and transferability of prompt representations.
Technical Breakthrough
To address these challenges, Virgilqiu's team introduces TPGC (Task-Prior and Structure-Prior based Graph Prompt Learning), a framework that explicitly models the synergy between task priors and structural priors. The key innovations include:
- Task-Prior Injection Module: Conducts short-term homologous multi-task pre-training on an auxiliary graph, allowing prompt initialization to inherit optimization preferences associated with multiple pretext tasks.
- Structure-Prior Injection Module: Extracts transferable global structural context from the auxiliary graph and converts it into layer-wise prompt vectors by aggregating structurally informative node embeddings.
Experimental Results
Experiments on six mainstream benchmarks covering node classification and graph classification demonstrate that TPGC achieves superior performance in few-shot settings:
- Performance Improvement: TPGC outperforms existing state-of-the-art methods in terms of accuracy across all benchmarks.
- Efficiency Optimization: TPGC requires fewer downstream tunable parameters and has lower runtime, significantly improving training efficiency.
Industry Impact and Developer Recommendations
- Advancing Graph Model Development: TPGC provides a new approach to adapting graph models in low-resource scenarios, which could drive further research and applications in the field of Graph Neural Networks (GNN).
- Developer Tool Support: The framework's code is open-source, allowing developers to build upon TPGC for customized applications and optimizations.
- Expanding Application Scenarios: TPGC holds promise for applications in bioinformatics, social network analysis, and recommendation systems, among others.
Future Directions
In the future, TPGC can be further refined by enhancing the synergy between task priors and structural priors and exploring its applications on larger-scale graph data. Additionally, researchers can integrate TPGC with other advanced technologies, such as self-supervised learning and meta-learning, to further boost the performance of graph models.
— END —Source: ArXiv Machine Learning (cs.LG) (2026-09-02)
Tags: #Graph Neural Networks #Multi-Task Learning #Pre-Training Models #Few-Shot Learning #Open-Source Framework
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments