Datasets for Link-Level Problems
Bitcoin-OTC (Link Classification)
The dataset Bitcoin-OTC is a who-trust-whom network, where nodes are users of a bitcoin-trading platform. Each directed edge has an integer rating between −10 to 10 and a timestamp. The ratings are divided into 6 groups. Two of them are used separately for Task-IL and accumulated for Class-IL. For Time-IL, we formulate 7 tasks using the timestamps, where the signs of the ratings are used as binary classes. Since there is no external node feature, we use in-degrees and out-degrees as node features.
Statistics:
Nodes: 5,881
Edges: 35,592
Number of Classes: 21
Supported Incremental Settings:
Task-IL with 3 tasks
Class-IL with 3 tasks
Time-IL with 7 tasks
Citing:
@inproceedings{kumar2016edge,
title={Edge weight prediction in weighted signed networks},
author={Kumar, Srijan and Spezzano, Francesca and Subrahmanian, VS and Faloutsos, Christos},
booktitle={ICDM},
year={2016},
}
@inproceedings{kumar2018rev2,
title={Rev2: Fraudulent user prediction in rating platforms},
author={Kumar, Srijan and Hooi, Bryan and Makhija, Disha and Kumar, Mohit and Faloutsos, Christos and Subrahmanian, VS},
booktitle={WSDM},
year={2018},
}
Wiki-CS (Link Prediction)
Wiki-CS is a hyperlink network between computer science articles. Each article has a label indicating one of the 10 subfields that it belongs to. For Domain-IL, the node labels are used as domains, and specifically, the edges are divide into 54 groups, according to the labels of their endpoints. According to the original paper, the node features are computed as the average of pretrained GloVe embeddings.
Statistics:
Nodes: 11,701
Edges: 431,726
Number of Classes: 2
Number of Node Features: 300
Supported Incremental Settings:
Domain-IL with 54 tasks
Citing:
@article{mernyei2020wiki,
title={Wiki-CS: A Wikipedia-Based Benchmark for Graph Neural Networks},
author={Mernyei, P{\'e}ter and Cangea, C{\u{a}}t{\u{a}}lina},
journal={arXiv preprint arXiv:2007.02901},
year={2020}
}
ogbl-collab (Link Prediction)
ogbl-collab s a co-authorship network, where nodes are authors. We use publication years to form 50 groups for the Time-IL setting. Due to the imbalance in the publication years of the considered papers, we constructed the first task using the paper published before the year 1971. For each subsequent i-th task, we used the papers published in the year (1969 + i).
Statistics:
Nodes: 235,868
Edges: 1,285,465
Number of Classes: 2
Number of Node Features: 128
Supported Incremental Settings:
Time-IL with 50 tasks
Citing:
@inproceedings{hu2020open,
title={Open graph benchmark: datasets for machine learning on graphs},
author={Hu, Weihua and Fey, Matthias and Zitnik, Marinka and Dong, Yuxiao and Ren, Hongyu and Liu, Bowen and Catasta, Michele and Leskovec, Jure},
booktitle={NeurIPS},
year={2020}
}
@article{wang2020microsoft,
title={Microsoft academic graph: When experts are not enough},
author={Wang, Kuansan and Shen, Zhihong and Huang, Chiyuan and Wu, Chieh-Han and Dong, Yuxiao and Kanakia, Anshul},
journal={Quantitative Science Studies},
volume={1},
number={1},
pages={396--413},
year={2020}
}
Facebook (Link Prediction)
Facebook is a social network, where nodes are pages of Facebook, and its class belongs to one among 8 categories. Edges indicate mutual likes among the pages, and they exist only between pages within the same category. We divide the edges into 8 groups, according to the labels of their endpoints. We make the graph to be undirected, and since there is no external node feature, we use degrees as node features.
Statistics:
Nodes: 134,833
Edges: 1,380,293
Number of Classes: 2
Number of Node Features: 1
Supported Incremental Settings:
Domain-IL with 8 tasks
Citing:
@inproceedings{rozemberczki2019gemsec,
title={GEMSEC: Graph Embedding with Self Clustering},
author={Rozemberczki, Benedek and Davies, Ryan and Sarkar, Rik and Sutton, Charles},
booktitle={ASONAM},
year={2019},
}
Ask-Ubuntu (Link Prediction)
Nodes in Ask-Ubuntu are users of askubuntu, and edges indicate there is interaction between the users. The edges are divided into 69 groups according to the timestamps for Time-IL. Specifically, we used the interactions occurring within the same month to form each task. We make the graph to be undirected, and since there is no external node feature, we use degrees as node features.
Statistics:
Nodes: 159,313
Edges: 507,988
Number of Classes: 2
Number of Node Features: 1
Supported Incremental Settings:
Time-IL with 69 tasks
Citing:
@inproceedings{paranjape2017motifs,
title={Motifs in temporal networks},
author={Paranjape, Ashwin and Benson, Austin R and Leskovec, Jure},
booktitle={Proceedings of the tenth ACM international conference on web search and data mining},
pages={601--610},
year={2017}
}
Gowalla (Link Prediction)
Gowalla consists of check-in history from a location-based social networking platform where users share their locations through check-ins. Each node represents either a user or a location, and each edge represents a user’s check-in at a location. For Time-IL, we organize 10 tasks chronologically based on check-in timestamps. This scenario and the next one (i.e., MovieLens) are directly related to personalized recommendation systems (spec., movie and POI recommendations), which are essential for helping users find relevant options among numerous candidates.
Statistics:
Nodes: 70,839
Edges: 1,027,370
Number of Classes: 2
Number of Node Features: 2
Supported Incremental Settings:
Time-IL with 10 tasks
Citing:
@inproceedings{liang2016modeling,
title={Modeling user exposure in recommendation},
author={Liang, Dawen and Charlin, Laurent and McInerney, James and Blei, David M},
booktitle={Proceedings of the 25th international conference on World Wide Web},
pages={951--961},
doi = {10.48550/arXiv.1510.07025},
year={2016}
}
@inproceedings{wang2019neural,
title={Neural graph collaborative filtering},
author={Wang, Xiang and He, Xiangnan and Wang, Meng and Feng, Fuli and Chua, Tat-Seng},
booktitle={Proceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval},
pages={165--174},
year={2019}
}
MovieLens (Link Prediction)
MovieLens is a movie-rating dataset. We convert it into a graph where nodes represent either users or movies. An edge is created between a user and a movie if and only if the user gives the movie a rating of 4 or higher. For Time-IL, we organize 10 tasks chronologically based on the rating timestamps. This scenario and the previous one (i.e., Gowalla) are directly related to personalized recommendation systems (spec., movie and POI recommendations), which are essential for helping users find relevant options among numerous candidates.
Statistics:
Nodes: 9,992
Edges: 575,281
Number of Classes: 2
Number of Node Features: 42
Supported Incremental Settings:
Time-IL with 10 tasks
Citing:
@article{harper2015movielens,
title={The movielens datasets: History and context},
author={Harper, F Maxwell and Konstan, Joseph A},
journal={Acm transactions on interactive intelligent systems (tiis)},
volume={5},
number={4},
pages={1--19},
year={2015},
doi = {10.1145/2827872},
publisher={Acm New York, NY, USA}
}