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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Semi-supervised Graph Anomaly Detection via Robust Homophily Learning
Authors: GUOGUO AI, Hezhe Qiao, Hui Yan, Guansong Pang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on eight real-world GAD datasets show that RHO can effectively learn varying, often under-represented, homophily in the small labeled node set and substantially outperforms state-of-the-art competing methods. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Nanjing University of Science and Technology 2School of Computing and Information Systems, Singapore Management University EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 RHO: Semi-supervised Graph Anomaly Detection via Robust Homophily Learning |
| Open Source Code | Yes | Code is available at https://github.com/mala-lab/RHO. |
| Open Datasets | Yes | We evaluate RHO on eight real-world GAD datasets of diverse size from different domains, including social networks Reddit [18] and Questions [35], co-review network Amazon [10], co-purchase network Photo [28], collaboration network Tolokers [28], financial networks T-Finance [41], Elliptic [43], and DGraph [15]. See App. B for more details about the datasets. |
| Dataset Splits | Yes | Following previous work [38], we randomly sample R% of the normal nodes as labeled data for training, where R {5, 10, 15, 20}. |
| Hardware Specification | Yes | The RHO model is implemented in Py Torch 2.0.0 with Python 3.8 and executed on Ge Force RTX 3090 GPU (24 GB). |
| Software Dependencies | Yes | The RHO model is implemented in Py Torch 2.0.0 with Python 3.8 and executed on Ge Force RTX 3090 GPU (24 GB). |
| Experiment Setup | Yes | RHO is trained using the Adam optimizer [16] with a weight decay of 5e 5. We set the default learning rate to 5e 3. Nevertheless, owing to the variations in edge density across different graphs, we find that models trained on sparser graphs are generally more sensitive to large learning rates and thus benefit from smaller ones to ensure stable convergence. Therefore, we use a learning rate of 5e 4 for the Elliptic and Question datasets, and further decreased to 5e 6 for the extremely sparse dataset, DGraph. The hyperparameter α is set to 1.0 for all datasets except the small datasets Reddit and Photo, which require less regularization. It is set to 0.1 in these two datasets. |