openbmb/factnet_factsense

FactSense Dataset

Overview

FactSense is the linguistic layer of FactNet that provides multilingual, natural language expressions of facts extracted from Wikipedia pages. Each FactSense instance represents a FactStatement realized in natural text with provenance information.

• 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:

• factsense_id: Unique identifier for the FactSense instance
• belongs_to_statement_id: ID of the associated FactStatement
• language: Language code (e.g., 'en', 'zh')
• page_id: Wikipedia page ID
• page_title: Title of the Wikipedia page
• page_namespace: Wikipedia namespace
• match_type: Type of match (sitelink, sitelink_pagelink, label_based, etc.)
• sentence: Text snippet containing the fact mention (limited to 1000 chars)
• sentence_index: Position of the sentence in the paragraph/page
• confidence: Match confidence score (0.5-0.9)
• subject_qid: Subject entity QID
• property_pid: Property PID
• value_qid: Value entity QID (if applicable)
• subject_label: Label used for matching
• value_label: Value label used for matching
• extraction_method: Extraction strategy version
• extraction_ts: Extraction timestamp

Usage

FactSense instances provide natural language evidence for facts across multiple languages, enabling applications in multilingual fact checking, knowledge-grounded generation, and cross-lingual information retrieval.

License

This dataset contains text derived from 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}
}