Wiki-CS: A Wikipedia-Based Benchmark for Graph Neural Networks
By Péter Mernyei, Cătălina Cangea. arXiv.org, 2020
374 citations recorded by the source corpus.
We present Wiki-CS, a novel dataset derived from Wikipedia for benchmarking Graph Neural Networks. The dataset consists of nodes corresponding to Computer Science articles, with edges based on hyperlinks and 10 classes representing different branches of the field. We use the dataset to evaluate semi-supervised node classification and single-relation link prediction models. Our experiments show that these methods perform well on a new domain, with structural properties different from earlier benchmarks. The dataset is publicly available, along with the implementation of the data pipeline and the benchmark experiments, at this https URL .
View Paper: https://www.semanticscholar.org/paper/8dd90b0648da4cb0e1f18ba1f00b158a157d3631
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
- Yoshua Bengio
- Vijay Prakash Dwivedi
- Shirui Pan
- Philip S. Yu
- Qiang Liu
- Yixin Liu
- Sergey Levine
- Dacheng Tao
Citations in the corpus
- Large-Scale Representation Learning on Graphs via Bootstrapping (2021, unknown): “Note that these are challenging medium-scale datasets specifically proposed for rigorous evaluation of semi-supervised node classification methods (Mernyei & Cangea, 2020; Shchur et al., 2018), but are almost all small enough that constrastive approaches such as GRACE (Zhu et al., 2020b) idealized…”
- Combining Label Propagation and Simple Models Out-performs Graph Neural Networks (2020, methodology): “For wikiCS, we use APPNP as reported by Mernyei & Cangea (2020).”
- Graph Self-Supervised Learning: A Survey (2021, methodology): “In Wiki-CS [135], nodes represent papers about computer science, edges represent the hyperlinks between papers, and node features are the mean vectors of GloVe word embeddings of articles.”
- Graph Contrastive Learning with Adaptive Augmentation (2020, methodology): “Wiki-CS has dense numerical features. While the other four datasets only contain sparse one-hot features. For the Wiki-CS dataset, we evaluate the models on the public splits shipped with the dataset [25]. Regarding the other four co-coauthor and co-purchase datasets, since they have no public split”
- Benchmarking Graph Neural Networks (2023, methodology): “WikiCS is a node classification dataset based on an extracted subset of Wikipedia’s Computer Science articles (Mernyei and Cangea, 2020).”
- How to Find Your Friendly Neighborhood: Graph Attention Design with Self-Supervision (2022, methodology): “Wiki-CS The Wiki-CS dataset is computer science related page networks in Wikipedia (Mernyei & Cangea, 2020).”
- Bootstrapped Representation Learning on Graphs (2021, methodology): “We first evaluate our method on a set of 5 recent real-world datasets (Mernyei & Cangea, 2020; Shchur et al., 2018) proposed for evaluating semi-supervised node classification — WikiCS, 1 Amazon-Computers, Amazon-Photos, Coauthor-CS, Coauthor-Physics 2 — in the transductive setting and show that our…”