CHoE: Cross-Domain Heterogeneous Graph Prompt Learning via Structure-Conditioned Experts

2026-05-15

AI Summary

Existing heterogeneous graph prompt learning methods suffer significant performance degradation in cross-domain scenarios due to distribution shifts. To address this challenge, this work proposes CHoE, the first approach to integrate structure-conditioned mixture-of-experts into cross-domain heterogeneous graph prompt learning. During pretraining, CHoE learns structure-aware expert modules; during prompt tuning, it dynamically selects experts via a structure-aware routing mechanism coupled with load balancing, and performs downstream prediction through prompt-driven multi-view semantic fusion. This design enables adaptive modeling of diverse metapath-based views, consistently outperforming existing baselines under few-shot cross-domain settings and substantially enhancing downstream task performance.

Abstract

Heterogeneous Graph Prompt Learning (HGPL)has emerged as a promising paradigm for bridging the gap between the objectives of pre-training foundation models and their downstream applications in heterogeneous graph settings. However, existing HGPL methods are primarily designed for in-domain scenarios, whereas real-world deployments often span multiple domains, and the data used for pre-training and downstream tasks may originate from different distributions. Consequently, the applicability of current HGPL approaches is limited to in-domain settings, and their performance typically degrades when application domains shift. To address this serious limitation, we develop CHoE, a cross-domain HGPL method built upon an expert network. During pre-training, we introduce and train structure-conditioned experts, and during prompt tuning, we adopt a structure-aware expert routing and load balancing mechanism to select structurally compatible experts for each meta-path view. In addition, we design a prompt-based semantic fusion module to integrate representations across multiple views for downstream prediction. Extensive experiments show that CHoE consistently improves performance in few-shot cross-domain applications, outperforming all baseline approaches.

Problem

• Cross-Domain
• Heterogeneous Graph
• Prompt Learning
• Domain Shift
• Foundation Models

Innovation

• Cross-Domain
• Heterogeneous Graph
• Prompt Learning
• Structure-Conditioned Experts
• Expert Routing