The data for Kernel Graph Attention Network for Fact Verification. Paper.

Can be found at Google Drive.

• The fact verification shared task contains three steps: Document Retrieval, Sentence Retrival and Fact Verification.
• We use the same document retrival with GEAR. Only the sentence retrieval part is different.

• The generate data format is same with the official data. Only evidence field is different.

{
    "id": 62037,
    "label": "SUPPORTS",
    "claim": "Oliver Reed was a film actor.",
    "evidence": [
        ...
    ]
}

• all_train.json; all_dev.json; all_test.json contains all sentences from retrieved document set.
  • Data format:

   {"evidence": [["Colin_Kaepernick", 0, "Colin Rand Kaepernick LRB LSB ` k\u00e6p\u0259rn\u026ak RSB ; born November 3 , 1987 RRB is an American football quarterback who is currently a free agent .", 0]]
  

   For evidence filed, [DocumentName(WikiTitle), evidence_id, evidence_content, golden/pseudo flag].
  

   Note that for the testing file, no golden label is provided. Hence the golden/pseudo flag is always set to 0.
  

• Retrieval results for claim verification.
  • bert_train.json; bert_dev.json; bert_test.json are the data for claim verification with BERT based retrieval.
  • esim_train.json; esim_dev.json; esim_test.json are the data for claim verification with ESIM based retrieval.
  • Data format:

   {"evidence": [["Colin_Kaepernick", 6, "He remained the team `s starting quarterback for the rest of the season and went on to lead the 49ers to their first Super Bowl appearance since 1994 , losing to the Baltimore Ravens .", 0.9736882448196411]]
  

   For evidence filed, [DocumentName(WikiTitle), evidence_id, evidence_content, retrieval_score]
  

• golden_train.json; golden_dev.json contain all golden sentences
  • Data format:

   {"evidence": [["Andrew_Kevin_Walker", 0, "Andrew Kevin Walker LRB born August 14 , 1964 RRB is an American BAFTA nominated screenwriter ."]]}
  

   For evidence filed, [DocumentName(WikiTitle), evidence_id, evidence_content]
  

• Process data for pairwise training.
  • run bash process.sh.