.png){kind=icon}
.png){kind=icon}
Return
MultiHGPT: Multi-task heterogeneous graph prompt tuning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ipm.2025.104236.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recently, the “pre-train, prompt” paradigm has emerged as a promising alternative to the “pre-train, fine-tune” approach, potentially bridging the extensive gap between pre-training and downstream stages in graph domains. Nonetheless, existing graph prompting techniques predominantly cater to single tasks within homogeneous graph structures, neglecting the heterogeneity and task disparity that is prevalent in real-world applications. In this paper, we propose MultiHGPT, a prompting method designed for multi-task heterogeneous graph pre-training and prompt tuning, to effectively mitigate the disparities caused by heterogeneity between pre-training and various downstream tasks. Specifically, we first integrate different level tasks into the same task subspace by an unified multi-task framework. Then to harness the rich semantic information inherent in heterogeneous graphs, we achieve multi-view decomposition of heterogeneous graphs through graph template construction. Moreover, by designing a novel prompting function for downstream multi-tasks, we bridge the gaps caused by heterogeneity differences between various tasks. Finally, we conduct experiments on three publicly available datasets under a few-shot setting, targeting downstream tasks at the node, edge, and graph levels. The predictions are made by selecting the class with the highest similarity to the class prototype. The experimental results validates the effectiveness of our method, achieving an average improvement of 9% over comparative methods.
Journal
I
IF:
6.9
Papers:
5.2K
Citations:
1.4W
Organization
No organization information available