Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
On the Benefits of Attribute-Driven Graph Domain Adaptation
Authors: Ruiyi Fang, Bingheng Li, zhao kang, Qiuhao Zeng, Nima Hosseini Dashtbayaz, Ruizhi Pu, Charles Ling, Boyu Wang
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on a variety of benchmark verify the effectiveness of our method. [...] 5 EXPERIMENT 5.1 DATASETS [...] 5.3 EXPERIMENTAL SETUP [...] 5.4 CROSS-NETWORK NODE CLASSIFICATION RESULTS [...] 5.5 ABLATION STUDY [...] 5.6 PARAMETER ANALYSIS |
| Researcher Affiliation | Academia | 1 Western University 2 Michigan State University 3 Vector Institute 4 University of Electronic Science and Technology of China EMAIL EMAIL EMAIL EMAIL |
| Pseudocode | Yes | C DESCRIPTION OF ALGORITHM GAA Algorithm 1: The proposed algorithm GAA Input: Source node feature matrix XS; source original graph adjacency matrix AS; Target node feature matrix XT ; Target original graph adjacency matrix AT source node label matrix Y S; maximum number of iterations η Compute the feature graph topological structure ˆ AS and ˆ AT according to XS and XT by running k NN algorithm. for it = 1 to η do ZS = GCN(AS, XS) ZS f = GCN( ˆAS, XS)// embedding of source graph ZT = GCN(AT , XT ) ZT f = GCN( ˆAT , XT )// embedding of target graph ZS and ZS f through cross-view similarity matrix refinement to get SS. ZT and ZT f through cross-view similarity matrix refinementto get ST . Attribute-Driven domain adaptive between SS and ST // adaptive in two views Domain Adaptive Learning between ZS and ZT ˆy S i constrained byy S i and ˆy T i constrained byˆy T i Calculate the overall loss with Eq.(15) Update all parameters of the framework according to the overall loss end Predict the labels of target graph nodes based on the trained framework. Output: Classification result ˆY T |
| Open Source Code | No | The paper does not explicitly provide a link to source code, nor does it state that the code for the described methodology is being released or is available in supplementary materials. |
| Open Datasets | Yes | To prove the superiority of our work on domain adaptation node classification tasks, we evaluate it on four types of datasets, including Airport dataset (Ribeiro et al., 2017), Citation dataset (Wu et al., 2020), Social dataset (Liu et al., 2024a) and Blog dataset (Li et al., 2015). [...] For a comprehensive overview of these datasets, please refer to Tab 5. |
| Dataset Splits | No | With a partially labeled node set V0 V, our objective in the node classification problem is to learn a model h : RN d GN RN C from a family of classifiers H that can predict the labels for the remaining unlabeled nodes. [...] We repeatedly train and test our model for five times with the same partition of dataset and then report the average of ACC. |
| Hardware Specification | Yes | The experiments are implemented in the Py Torch platform using an Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz, and Ge Force RTX A5000 24G GPU. |
| Software Dependencies | No | The experiments are implemented in the Py Torch platform using an Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz, and Ge Force RTX A5000 24G GPU. Technically, two layers GCN is built and we train our model by utilizing the Adam (Kingma & Ba, 2015) optimizer with learning rate ranging from 0.0001 to 0.0005. |
| Experiment Setup | Yes | The experiments are implemented in the Py Torch platform using an Intel(R) Xeon(R) Silver 4210R CPU @ 2.40GHz, and Ge Force RTX A5000 24G GPU. Technically, two layers GCN is built and we train our model by utilizing the Adam (Kingma & Ba, 2015) optimizer with learning rate ranging from 0.0001 to 0.0005. In order to prevent over-fitting, we set the dropout rate to 0.5. In addition, we set weight decay {1e 4, , 5e 3} and k {1, , 10} for k NN graph. For fairness, we use the same parameter settings for all the cross-domain node classification methods in our experiment, except for some special cases. For GCN, UDA-GCN, and JHGDA the GCNs of both the source and target networks contain two hidden layers (L = 2) with structure as 128 16. The dropout rate for each GCN layer is set to 0.3. We repeatedly train and test our model for five times with the same partition of dataset and then report the average of ACC. [...] α, β, and τ are chosen from the set {0.005, 0.01, 0.1, 0.5, 1, 5}. These values provide flexibility for adjusting the relative importance of different loss terms. k (the number of neighbors for k-NN graph construction) is typically k {1, , 10}. [...] Table 5: Experiment hyperparameter setting Value. |