Complete-Tree Space Favors Data-Efficient Link Prediction

Authors: Chi Gao, Lukai Li, Yancheng Zhou, Shangqi Guo

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments verify the data efficiency of CT space over other spaces. Moreover, leaf matching outperforms the state-of-the-art graph transformer in data-scarce scenarios while exhibiting excellent scalability. We experiment on six datasets, including the classic Cora, Pubmed, Citeseer (Yang et al., 2016; Li et al., 2024), the large-scale ogbl-collab, ogbl-ppa (Hu et al., 2020), and the temporal ICEWS18 (Liu et al., 2022).
Researcher Affiliation Academia 1Center for Brain-Inspired Computing Research, Department of Precision Instrument, Tsinghua University, Beijing, China 2Weixian College, Tsinghua University, Beijing, China. Correspondence to: Shangqi Guo <shangqi EMAIL>.
Pseudocode Yes Algorithm 1 Greedy-Navigation-based Optimization
Open Source Code Yes The code is available at: https://github.com/Kevin Gao7/Leaf Matching.
Open Datasets Yes We experiment on six datasets, including the classic Cora, Pubmed, Citeseer (Yang et al., 2016; Li et al., 2024), the large-scale ogbl-collab, ogbl-ppa (Hu et al., 2020), and the temporal ICEWS18 (Liu et al., 2022).
Dataset Splits Yes For data splitting and the negative ratio in testing, we adhere to the settings established by Cog DL (Cen et al., 2023).
Hardware Specification No No specific hardware details (GPU/CPU models, memory, etc.) are provided in the paper. The paper only vaguely mentions 'excessive memory (on the order of terabytes)' in the context of an out-of-memory error for SEAL, but does not specify the hardware used for their own experiments.
Software Dependencies No The paper mentions using 'official implementation of node2vec, node2ket, and poincare' and 'official implementations of LPFormer, GAE, and SEAL provided in Hea RT (Li et al., 2023)', but it does not specify any software versions (e.g., Python, PyTorch, specific library versions).
Experiment Setup Yes For leaf matching, in Cora, Citeseer and Pubmed, we set h = 12, γ = 3, α = 8, c = 1, λ = 1, re = 0.8, training it with batch size 256 on 50 epochs. For ICEWS18, we set h = 14, γ = 3, α = 33, c = 3, λ = 1, re = 0.8, training it with batch size 32 on 50 epochs. For ogbl-collab, we set h = 16, γ = 3, α = 57, c = 3, λ = 1, re = 0.8, training it with batch size 256 on 50 epochs.