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..
One Prompt Fits All: Universal Graph Adaptation for Pretrained Models
Authors: Yongqi Huang, Jitao Zhao, Dongxiao He, Xiaobao Wang, Yawen Li, Yuxiao Huang, Di Jin, Zhiyong Feng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that our method can effectively integrate with various pretrained models and achieve strong performance across in-domain and cross-domain scenarios. We evaluate the effectiveness of Uni Prompt 1 using nine node classification datasets, including three homophilic datasets Cora [34], Cite Seer [34] and Pub Med [34], and six heterophilic datasets Cornell [35], Texas [35], Wisconsin [35], Chameleon [35], Actor [35] and Squirrel [35]. For in-domain settings, we use DGI [28], GRACE [29], and Graph MAE [30] as pretrained models, and we compare our method with two baseline tuning methods, and seven classic and state-of-the-art GPL methods... |
| Researcher Affiliation | Academia | 1College of Intelligence and Computing, Tianjin University, 2School of Economics and Management, Beijing University of Posts and Telecommunications, 3Department of Data Science, George Washington University, 1EMAIL EMAIL, EMAIL |
| Pseudocode | No | The paper includes mathematical formulations and definitions of functions (e.g., Theorem 4.1, Proposition 4.1), but it does not contain a distinct section or figure explicitly labeled as "Pseudocode" or "Algorithm" with structured steps. |
| Open Source Code | Yes | 1Code is available at: https://github.com/hedongxiao-tju/Uni Prompt |
| Open Datasets | Yes | We evaluate the effectiveness of Uni Prompt 1 using nine node classification datasets, including three homophilic datasets Cora [34], Cite Seer [34] and Pub Med [34], and six heterophilic datasets Cornell [35], Texas [35], Wisconsin [35], Chameleon [35], Actor [35] and Squirrel [35]. |
| Dataset Splits | Yes | Following the Pro G [2] benchmark settings, we conduct k-shot sampling evaluations under both in-domain and cross-domain settings with k {1, 3, 5}. To ensure performance reliability, we perform 20 repeated runs for each of 5 fixed random seeds = { 42, 12345, 23344, 38108, 39788 }, reporting averaged results over 100 total trials. |
| Hardware Specification | Yes | All of the experiments are conducted on a server with Xeon(R) Platinum 8352V CPU, 90GB of memory, an RTX 4090 graphics card, and 24GB of video memory. |
| Software Dependencies | No | The paper mentions using GCN, GAT, DGI, GRACE, Graph MAE as backbones or pretrained models and refers to GitHub links for these, but it does not provide specific version numbers for software libraries like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | We train for 2000 epochs with early stopping (patience=20). Following the Pro G [2] benchmark settings, we conduct k-shot sampling evaluations under both in-domain and cross-domain settings with k {1, 3, 5}. To ensure performance reliability, we perform 20 repeated runs for each of 5 fixed random seeds = { 42, 12345, 23344, 38108, 39788 }, reporting averaged results over 100 total trials. ... We conduct extensive experiments to explore the impact of various hyperparameters on the performance of our model, as shown in Table 11, ensuring that our approach achieves robust and consistent results across diverse settings. Table 11 provides 'Hyperparameter settings of Uni Prompt for 1-shot, 3-shot, and 5-shot scenarios across different pretrained models' including 'up_lr', 'down_lr', 'k', and 'τ' values for each dataset and shot setting. |