DELFT: Complex Factoid QA with a Free-Text Knowledge Graph

DELFT is a factoid question answering system that combines the nuance and depth of knowledge graph QA with the broader coverage of free text. It builds a free-text knowledge graph from Wikipedia — entities as nodes, co-occurrence sentences as edges — and uses a graph neural network to reason over it.

Approach

DELFT constructs dense semantic graphs and employs graph neural networks to combine node evidence through edge sentences to determine answers. This hybrid approach outperforms pure machine reading and BERT-based baselines on entity-rich questions.

Wikipedia Graph Data

Two versions are provided:

Version Description Download
Original links Wikipedia article hyperlinks as edges enwiki_links_anchor_1101.zip
TagMe entities Entity annotations via TagME as edges wiki_anchor_1101.zip

Each entry contains: page id, title, text, anchored entities (from TagMe or article links).

Resources

Citation

@inproceedings{Zhao:Xiong:Qian:Boyd-Graber-2020,
  Title  = {Complex Factoid Question Answering with a Free-Text Knowledge Graph},
  Author = {Chen Zhao and Chenyan Xiong and Xin Qian and Jordan Boyd-Graber},
  Booktitle = {The Web Conference},
  Year   = {2020}
}

Contact

Chen Zhao — chenz@cs.umd.edu