📊 Datasets — Awesome Graph Learning

The CiteSeer dataset consists of 3312 scientific publications classified into one of six classes. The citation network consists of 4732 links. Each publication in the dataset is described by a 0/1-valued word vector indicating the absence/presence of the corresponding word from the dictionary. The dictionary consists of 3703 unique words. The README file in the dataset provides more details.

A citation-network benchmark for node classification (papers linked by citations, classified by topic).

A citation-network benchmark of biomedical papers for node classification.

A collection of large-scale, realistic graph datasets with a standardized evaluation for graph machine learning.

134k small organic molecules with computed quantum-chemical properties, for molecular-property prediction.

The Open Graph Benchmark (OGB) is a collection of large-scale node classification datasets used to evaluate the performance of graph neural networks (GNNs).

The 'Reddit' dataset is a real-world temporal graph used to evaluate the performance of conformal prediction methods for graph neural networks in capturing uncertainties in dynamic settings.

The 'ogbn-papers-100M' dataset contains a large-scale graph of academic papers and is used to evaluate the performance of graph neural networks (GNNs) in representation learning.

WN-18RR is a benchmark dataset used to evaluate knowledge graph completion methods, containing a subset of the WordNet lexical database with relationships between entities.

Dataset Card for ZINC Dataset Summary The ZINC dataset is a "curated collection of commercially available chemical compounds prepared especially for virtual screening" (Wikipedia). Supported Tasks and Leaderboards ZINC should be used for molecular property prediction (aiming to predict the constrained solubility of the molecules), a graph regression task. The score used is the MAE. The associated leaderboard is here: Papers with code leaderboard.… See the full description on the dataset page: https://huggingface.co/datasets/graphs-datasets/ZINC.

FB-15K-237 is a benchmark dataset for knowledge graph completion that contains a collection of entities and relationships used to evaluate the performance of various models in inferring missing links within knowledge graphs.

The PPI dataset contains protein-protein interaction data and is used to evaluate semi-supervised node classification algorithms.

MovieLens is a dataset used to evaluate recommendation systems, containing user-item interactions that facilitate the assessment of algorithms' performance in predicting user preferences.

The 'PCQM-4Mv-2' dataset is a benchmark for evaluating graph neural networks (GNNs) through the graph alignment problem, containing synthetic and real-world graph data used to assess model performance across varying task difficulties.

Audio-Centric Multimodal Benchmark (ACM) This dataset packages the ACM benchmark introduced by Efficient and High-Fidelity Omni Modality Retrieval. The paper is available at arXiv:2603.02098. Dataset repo: chuonghm/ACM ACM contains four HuggingFace subsets. Each subset uses a single test split so query and candidate tables can keep their natural schemas: composed_audio_retrieval_queries: flattened AT2A query rows. composed_audio_retrieval_candidates: AT2A audio candidate pool.… See the full description on the dataset page: https://huggingface.co/datasets/chuonghm/ACM.

The 'Actor' dataset is a benchmark used to evaluate the performance of graph neural networks (GNNs) on heterophilic graphs, where connected nodes have dissimilar features or labels.

The 'Bitcoin-Alpha' dataset is a benchmark that contains feature-sparse signed networks used to evaluate anomaly detection methods in Bitcoin trust systems.

The 'Bitcoin OTC' dataset contains historical data on over-the-counter Bitcoin trades and is used to evaluate link prediction methods in the context of temporal signed networks.

CIFAR-10 is a dataset containing 60,000 32x32 color images across 10 different classes, commonly used to evaluate the performance of machine learning models in image classification tasks.

The 'Coauthor CS' dataset contains co-authorship information in the computer science domain and is used to evaluate the effectiveness of unlearning strategies for graph neural networks.

The 'Co-author-Physics' dataset is a benchmark used to evaluate the performance of Graph Neural Networks by providing a collection of co-authorship relationships in the field of physics.

DBLP is a dataset that contains bibliographic information on computer science publications and is used to evaluate link prediction models in graph mining.

Freebase is a large-scale knowledge graph that contains structured information about entities and their relationships, and it is used to evaluate graph representation learning methods in the context of node classification tasks.

The MD-17 dataset is a benchmark that contains molecular dynamics data used to evaluate the performance of machine learning interatomic potentials.

Dataset Card for "mini_imagenet" More Information needed

Dataset Card for MNIST Dataset Summary The MNIST dataset consists of 70,000 28x28 black-and-white images of handwritten digits extracted from two NIST databases. There are 60,000 images in the training dataset and 10,000 images in the validation dataset, one class per digit so a total of 10 classes, with 7,000 images (6,000 train images and 1,000 test images) per class. Half of the image were drawn by Census Bureau employees and the other half by high school students… See the full description on the dataset page: https://huggingface.co/datasets/ylecun/mnist.

MoleculeNet Benchmark (website) MoleculeNet is a benchmark specially designed for testing machine learning methods of molecular properties. As we aim to facilitate the development of molecular machine learning method, this work curates a number of dataset collections, creates a suite of software that implements many known featurizations and previously proposed algorithms. All methods and datasets are integrated as parts of the open source DeepChem package(MIT license). MoleculeNet… See the full description on the dataset page: https://huggingface.co/datasets/katielink/moleculenet-benchmark.

NELL (Never-Ending Language Learning) is a dataset that contains a continuously growing knowledge base of facts extracted from the web, used to evaluate the robustness and generalization of models in graph-based learning tasks.

The NTU-RGB+D dataset is a benchmark that contains skeleton data for human actions and interactions, used to evaluate methods for recognizing both individual actions and human-human interactions.

An OGB citation-network node-classification benchmark over arXiv CS papers.

The Open Academic Graph is a large-scale dataset that contains academic papers and their citation relationships, used to evaluate the performance of graph neural networks in modeling graph-structured data.

The 'Planetoid' dataset is a benchmark used to evaluate link prediction tasks in graph learning, containing graph-structured data that facilitates the assessment of various heuristic learning methods.

Dataset Card for PROTEINS Dataset Summary The PROTEINS dataset is a medium molecular property prediction dataset. Supported Tasks and Leaderboards PROTEINS should be used for molecular property prediction (aiming to predict whether molecules are enzymes or not), a binary classification task. The score used is accuracy, using a 10-fold cross-validation. External Use PyGeometric To load in PyGeometric, do the following: from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/graphs-datasets/PROTEINS.

The Slashdot dataset contains user interactions and relationships within the Slashdot social news website and is used to evaluate link sign prediction in signed graph neural networks.

The 'Texas' dataset is a benchmark used to evaluate the performance of Graph Neural Networks (GNNs) on heterophilic graphs.

The 'Wisconsin' dataset is a benchmark used to evaluate node-level representation learning methods on heterophilic graphs, specifically assessing their ability to capture structural equivalence and hidden relationships among nodes.

The '11 benchmark' is a dataset used to evaluate the performance of Graph Neural Networks (GNNs) on various graph analytical tasks, particularly in distinguishing between homophilic and heterophilic linking patterns.

The '11 benchmark graphs' dataset contains a collection of diverse graph structures used to evaluate the effectiveness of graph contrastive learning methods.

The '12 regular and geometric graph benchmarks' dataset contains a variety of graph structures used to evaluate the performance of interpretable graph neural networks (XGNNs) in tasks such as graph classification.

The '13 benchmarks' dataset consists of various graph datasets used to evaluate the performance of graph neural networks, particularly in terms of accuracy on both homophilic and heterophilic graphs.

The '14 node classification benchmarks' consists of various datasets used to evaluate the performance of Graph Neural Networks (GNNs) in classifying nodes within graphs.

The '152 distinct graph datasets' refers to a collection of diverse graph datasets used to evaluate the scalability and generalizability of the GraphFM model, encompassing various domains such as molecules, citation networks, and product graphs, and totaling 7.4 million nodes and 189 million edges.

The '19 real-world networks' dataset contains various real-world network structures used to evaluate the performance of algorithms in predicting node importance, particularly through centrality measures like betweenness centrality.

The '1 synthetic' dataset/benchmark is used to evaluate the performance of the proposed Edge Splitting GNN (ES-GNN) framework in distinguishing between task-relevant and irrelevant edges in graph neural networks.

The '26 UEA benchmark datasets' consists of multivariate time series data used to evaluate the performance of classification algorithms in the context of multivariate time series classification (MTSC).

3D point cloud data consists of a set of points in three-dimensional space that represent the shape and structure of objects, and it is used to evaluate graph-based learning methods for tasks such as segmentation and classification.

The '3DPW' dataset is a benchmark that contains 3D human pose data used to evaluate human motion prediction models.

The "6 standard node classification benchmarks" refers to a set of datasets used to evaluate the performance of Graph Neural Networks (GNNs) in classifying nodes based on their features and neighborhood structures.

The '7 widely-used graph datasets' refers to a collection of benchmark datasets utilized to evaluate the performance of Graph Neural Networks (GNNs) and their susceptibility to the over-smoothing problem.

The '8 common real-world networks' dataset/benchmark contains a collection of diverse network structures used to evaluate the performance and degree bias of Graph Neural Networks (GNNs) in node classification tasks.

The '9 collected multi-label datasets' refers to a set of datasets used to evaluate the performance of various methods in multi-label node classification tasks, addressing the challenges posed by the scarcity of publicly available datasets in this area.

The '9 graph datasets' is a benchmark containing various graph data used to evaluate the performance of quantization-aware training methods for Graph Neural Networks.

The ADNI (Alzheimer's Disease Neuroimaging Initiative) dataset contains MRI data and associated demographic and genetic information from subjects, and it is used to evaluate models for Alzheimer's disease classification and causal analysis of disease progression.

The 'Alibaba mobile application' dataset is a real-world heterogeneous graph used to evaluate node classification, link prediction, and online A/B testing methodologies.

The Alibaba recommendation dataset is an industrial-sized dataset used to evaluate the performance of recommendation algorithms, particularly in the context of node classification tasks.

The 'AliExpress' dataset is a benchmark that contains structured data used to evaluate link prediction methods in recommendation systems, particularly focusing on the incorporation of edge attributes in large-scale sparse graphs.

The 'Alipay' dataset is a benchmark used to evaluate the performance of algorithms in graph learning, specifically in the context of graph-structured data related to user interactions and transactions within the Alipay platform.

The 'Amazon' dataset is a real-world fraud dataset used to evaluate the effectiveness of verification methods for graph neural networks against adversarial attacks.

The 'Amazon-2M' dataset contains 2 million nodes and 61 million edges, and it is used to evaluate the scalability and efficiency of the Cluster-GCN algorithm for training deep graph convolutional networks.

The 'Amazon Computers' dataset is a benchmark used to evaluate the performance of graph neural networks in the context of knowledge distillation and collaborative training.