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..
Pseudo-Riemannian Graph Transformer
Authors: Viet Quan Le, Cuong Viet Ta
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on diverse real-world graphs demonstrate that our model consistently outperforms other baselines in both node classification and link prediction tasks. |
| Researcher Affiliation | Academia | 1Department of Computer Science 2Human-Machine Interaction Laboratory VNU University of Engineering and Technology, Hanoi, Vietnam EMAIL |
| Pseudocode | Yes | In this section, we provide the pseudocode of our proposed framework to facilitate implementation for the reader. Algorithm 1 outlines the procedure of the space searching algorithm. [...] Algorithm 2 demonstrates the learning process of Q-GCN2. [...] Algorithm 3 describes the learning process of Q-GCT. |
| Open Source Code | Yes | Our implementation is available at the Git Hub repository https://github.com/quanlv9211/QGT.git. |
| Open Datasets | Yes | Datasets. We choose seven widely used graph datasets for evaluation, including four citation networks Cora (Mc Callum et al., 2000), Citeseer (Giles et al., 1998), Pubmed (Sen et al., 2008), and Arxiv (Hu et al., 2020); two social networks Penn94 and Twitch Gamers (Lim et al., 2021); and one transportation network Airport (Xiong et al., 2022). [...] Products and Vessels (Hu et al., 2020) |
| Dataset Splits | Yes | For the node classification task, we follow the standard dataset splits used by Hu et al. (2020); Xiong et al. (2022); Lim et al. (2021). [...] For the link prediction task, we evaluate performance on four datasets: Cora, Citeseer, Pubmed, and Airport. Each dataset s edges are split into 85%, 5%, and 10% for training, validation, and testing, respectively. |
| Hardware Specification | Yes | All the experiments are conducted on a GPU device NVIDIA Ge Force RTX 4090 with 24GB memory, except for the Products and Vessel datasets, which are run on an NVIDIA A100 GPU with 80GB of memory. |
| Software Dependencies | No | We implement our models by Pytorch and Geoopt tool 2. All the experiments are conducted on a GPU device NVIDIA Ge Force RTX 4090 with 24GB memory, except for the Products and Vessel datasets, which are run on an NVIDIA A100 GPU with 80GB of memory. |
| Experiment Setup | Yes | Hyperparameter. Besides the space and time dimensions of pseudo-hyperboloids Qp,q β , our models involves other hyperparameters, including learning rate lr, balance weight α, number of layers, activation function, dropout rate, and weight decay. The optimal hyperparameters are obtained by applying grid search strategy, where the ranges summarized in Table 12. [...] Each model is trained for up to 1000 epochs with early stopping applied. |