KG20C & KG20C-QA: Scholarly Knowledge Graph Benchmarks for Link Prediction and Question Answering

May 12, 2026 ยท View on GitHub

KG20C is a curated scholarly knowledge graph constructed from the Microsoft Academic Graph (MAG), covering authors, papers, venues, affiliations, domains, and citations from 20 top computer science conferences. It is designed as a benchmark dataset with standardized train/validation/test splits, following best practices similar to popular knowledge graph benchmarks such as WN18RR and FB15k-237.

KG20C-QA is a question answering benchmark built on top of KG20C. It provides both natural language questions and structured entity-relation queries on scholarly data. This design enables evaluation of both graph-based models (e.g., knowledge graph embeddings) and text-based models (e.g., large language models) under a unified benchmark setting.

KG20C provides reusable, extensible benchmarks for link prediction, and serves as a foundation for building derivative datasets for question answering and reasoning over scholarly data. For a complete description of the dataset construction, schema, and evaluation baselines, please refer to our papers. Note: This repository currently provides only the KG20C dataset. Other derivative datasets built upon KG20C for question answering and reasoning will be released in the future.

KG20C graph
Figure 1: Overview of the KG20C knowledge graph.

Dataset content (KG20C)

File format

All files are in tab-separated-values format, compatible with other popular benchmark datasets including WN18RR and FB15K-237. For example, train.txt includes "28674CFA author_in_affiliation 075CFC38", which denotes the author with id 28674CFA works in the affiliation with id 075CFC38.

The repo includes these files:

  • all_entity_info.txt contains id name type of all entities
  • all_relation_info.txt contains id of all relations
  • train.txt contains training triples of the form entity_1_id relation_id entity_2_id
  • valid.txt contains validation triples
  • test.txt contains test triples

Statistics

Data statistics of the KG20C knowledge graph:

AuthorPaperConferenceDomainAffiliation
8,6805,047201,923692
EntitiesRelationsTraining triplesValidation triplesTest triples
16,362548,2133,6703,724

License

The dataset is free to use for research purpose. For other uses, please follow Microsoft Academic Graph license.

How to cite

If you found the datasets or our work useful, please cite us.

For the KG20C & KG20C-QA datasets and baseline results, please cite:

  • Hung-Nghiep Tran and Atsuhiro Takasu. KG20C & KG20C-QA: Scholarly Knowledge Graph Benchmarks for Link Prediction and Question Answering. arXiv:2512.21799v2, 2025. doi: 10.48550/arXiv.2512.21799. 
    @misc{tran_kg20ckg20cqascholarly_2025,
  title = {{{KG20C}} \& {{KG20C-QA}}: {{Scholarly Knowledge Graph Benchmarks}} for {{Link Prediction}} and {{Question Answering}}},
  author = {Tran, Hung-Nghiep and Takasu, Atsuhiro},
  year = 2025,
  number = {arXiv:2512.21799},
  publisher = {arXiv},
  doi = {10.48550/arXiv.2512.21799},
  url = {http://arxiv.org/abs/2512.21799}
}

For the semantic query method and preliminary results, please cite:

  • Hung-Nghiep Tran. Multi-Relational Embedding for Knowledge Graph Representation and Analysis. PhD Dissertation, The Graduate University for Advanced Studies, SOKENDAI, Japan, 2020. 
    @phdthesis{tran_multirelationalembeddingknowledge_2020,
  address = {Japan},
  type = {{PhD} {Dissertation}},
  title = {Multi-{Relational} {Embedding} for {Knowledge} {Graph} {Representation} and {Analysis}},
  school = {The Graduate University for Advanced Studies, SOKENDAI},
  author = {Tran, Hung-Nghiep},
  year = {2020},
}
  • Hung-Nghiep Tran and Atsuhiro Takasu. Exploring Scholarly Data by Semantic Query on Knowledge Graph Embedding Space. In Proceedings of International Conference on Theory and Practice of Digital Libraries (TPDL), 2019. 
    @inproceedings{tran_exploringscholarlydata_2019,
  title = {Exploring {Scholarly} {Data} by {Semantic} {Query} on {Knowledge} {Graph} {Embedding} {Space}},
  booktitle = {Proceedings of the 23rd {International} {Conference} on {Theory} and {Practice} of {Digital} {Libraries}},
  author = {Tran, Hung-Nghiep and Takasu, Atsuhiro},
  year = {2019},
  pages = {154--162},
  url = {https://arxiv.org/abs/1909.08191},
}

For the MEI and MEIM baseline knowledge graph embedding models, please cite:

  • Hung-Nghiep Tran and Atsuhiro Takasu. Multi-Partition Embedding Interaction with Block Term Format for Knowledge Graph Completion. In Proceedings of the European Conference on Artificial Intelligence (ECAI), 2020. 
    @inproceedings{tran_multipartitionembeddinginteraction_2020,
  title = {Multi-{Partition} {Embedding} {Interaction} with {Block} {Term} {Format} for {Knowledge} {Graph} {Completion}},
  booktitle = {Proceedings of the {European} {Conference} on {Artificial} {Intelligence}},
  author = {Tran, Hung-Nghiep and Takasu, Atsuhiro},
  year = {2020},
  pages = {833--840},
  url = {https://arxiv.org/abs/2006.16365},
}
  • Hung-Nghiep Tran and Atsuhiro Takasu. MEIM: Multi-partition Embedding Interaction Beyond Block Term Format for Efficient and Expressive Link Prediction. In Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI), 2022. 
    @inproceedings{tran_meimmultipartitionembedding_2022,
  title = {{MEIM}: {Multi}-partition {Embedding} {Interaction} {Beyond} {Block} {Term} {Format} for {Efficient} and {Expressive} {Link} {Prediction}},
  booktitle = {Proceedings of the {Thirty}-{First} {International} {Joint} {Conference} on {Artificial} {Intelligence}},
  author = {Tran, Hung-Nghiep and Takasu, Atsuhiro},
  year = {2022},
  pages = {2262--2269},
  url = {https://arxiv.org/abs/2209.15597},
}

For the raw Microsoft Academic Graph dataset, please cite:

See also