DELTA: Dual Consistency Delving with Topological Uncertainty for Active Graph Domain Adaptation

TL;DR

This work tackles active graph domain adaptation by introducing DELTA, which learns two complementary topological views—edge-oriented and path-oriented subnetworks—to identify informative target nodes under a labeling budget. It combines subgraph-level topological uncertainty with a cross-domain discrepancy score to select high-value nodes in one round, improving target-domain performance while reducing annotation costs. Empirical results on multiple citation graphs show DELTA outperforming strong baselines and exhibiting strong robustness to backbone choices and hyperparameters, with analyses confirming the value of dual topology and uncertainty-driven selection. The approach offers a practical, scalable framework for cross-graph transfer learning where target labels are scarce, and it points to future work in open-set domain adaptation and annotation-efficient strategies using language models.

Abstract

Graph domain adaptation has recently enabled knowledge transfer across different graphs. However, without the semantic information on target graphs, the performance on target graphs is still far from satisfactory. To address the issue, we study the problem of active graph domain adaptation, which selects a small quantitative of informative nodes on the target graph for extra annotation. This problem is highly challenging due to the complicated topological relationships and the distribution discrepancy across graphs. In this paper, we propose a novel approach named Dual Consistency Delving with Topological Uncertainty (DELTA) for active graph domain adaptation. Our DELTA consists of an edge-oriented graph subnetwork and a path-oriented graph subnetwork, which can explore topological semantics from complementary perspectives. In particular, our edge-oriented graph subnetwork utilizes the message passing mechanism to learn neighborhood information, while our path-oriented graph subnetwork explores high-order relationships from sub-structures. To jointly learn from two subnetworks, we roughly select informative candidate nodes with the consideration of consistency across two subnetworks. Then, we aggregate local semantics from its K-hop subgraph based on node degrees for topological uncertainty estimation. To overcome potential distribution shifts, we compare target nodes and their corresponding source nodes for discrepancy scores as an additional component for fine selection. Extensive experiments on benchmark datasets demonstrate that DELTA outperforms various state-of-the-art approaches. The code implementation of DELTA is available at https://github.com/goose315/DELTA.

Figures (14)

• Figure 1: The problem setting of active learning for graph domain adaption. We have a fixed budget and aim to find the most informative nodes for data annotation.
• Figure 2: Overview of the proposed DELTA framework. DELTA first captures complementary topological semantics from the target graph via edge-oriented and path-oriented subnetworks. It then identifies candidate nodes with semantic inconsistencies between these subnetworks. DELTA computes topological uncertainty using degree weighting and K-hop subgraphs, and domain gap between the source and target graphs for informative node selection.
• Figure 3: Performance of proposed method and baseline methods by active learning budget, averaged from 5 different runs. The Macro-F1 scores are plotted.
• Figure 4: Performance comparison of ablation study.
• Figure 5: Performance of the proposed method by the consistency thresholds (left) and the number of hops for the K-hop subgraph (right), averaged from 5 different runs. In each sub-figure, the Macro-F1 and Micro-F1 scores are plotted. 50 nodes are selected for active learning.

Theorems & Definitions (2)

• Theorem 4.1
• proof