WikiQA Knowledge Graphs Dataset

This dataset contains Wikipedia articles paired with automatically generated knowledge graphs for question answering research.

Dataset Structure

• articles/ - Original Wikipedia articles in text format
• articles_*_kg/ - Knowledge graphs generated for different article subsets in JSON format
• aggregated/ - Final aggregated and deduplicated knowledge graphs and statistics
• *.csv - Question answering train/validation/test splits

Subsets

• articles_1/ - Single article (2006 Winter Olympics)
• articles_40k_ch/ - ~40k characters
• articles_400k_ch/ - ~400k character
• articles_4m_ch/ - ~4M character
• articles_20m_ch/ - ~20M character
• articles_w_context/ - Articles with additional context

Knowledge Graph Generation

Knowledge graphs were generated using the KG-Gen library with entity extraction, relation extraction, and graph clustering steps.

Citation

If you use this dataset in your research, please cite from https://arxiv.org/abs/2502.09956