openbmb/factnet_relations

FactNet Relations Dataset

Overview

The Synset Relations dataset contains rich semantic relationships between FactSynsets, enabling advanced reasoning and cross-lingual fact retrieval. These relations capture hypernymy, causality, temporality, geographic relationships, and other semantic connections between facts.

• Paper: https://arxiv.org/abs/2602.03417
• Github: https://github.com/yl-shen/factnet
• Dataset: https://huggingface.co/collections/openbmb/factnet

Dataset Format

The dataset contains parquet files with the following key fields:

• relation_id: Unique identifier for the relation
• source_synset_id: Source FactSynset ID
• target_synset_id: Target FactSynset ID
• relation_type: Type of relation (hypernym, causal, temporal, geographic, etc.)
• confidence: Confidence score for the relation
• evidence_statement_ids: FactStatements supporting this relation
• detection_method: Method used to detect the relation
• metadata: Additional relation-specific metadata

Relation Types

The dataset includes various relation types:

• equivalent: Semantically equivalent facts
• hypernym: Hierarchical relationships
• causal: Cause-effect relationships
• geographic_location/geographic_contains: Spatial relationships
• part_of/has_part: Part-whole relationships
• member_of: Membership relationships
• follows/followed_by: Temporal sequence
• influenced_by/influences: Influence relationships
• And many others including created_by, used_for, opposite_of, etc.

Usage

Synset Relations enable advanced applications like:

• Multi-hop reasoning across facts
• Causal and temporal inference
• Geographic and spatial reasoning
• Semantic similarity computation
• Hierarchical knowledge navigation

License

This dataset is derived from Wikidata and Wikipedia and is available under the CC BY-SA license.

Citation

@article{shen2026factnet,
  title={FactNet: A Billion-Scale Knowledge Graph for Multilingual Factual Grounding},
  author={Shen, Yingli and Lai, Wen and Zhou, Jie and Zhang, Xueren and Wang, Yudong and Luo, Kangyang and Wang, Shuo and Gao, Ge and Fraser, Alexander and Sun, Maosong},
  journal={arXiv preprint arXiv:2602.03417},
  year={2026}
}