Skip to content

We construct and introduce DIALFACT, a testing benchmark dataset crowd-annotated conversational claims, paired with pieces of evidence from Wikipedia.

License

Notifications You must be signed in to change notification settings

salesforce/DialFact

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

DialFact: A Benchmark for Fact-Checking in Dialogue

Authors: Prakhar Gupta, Jason Wu, Wenhao Liu and Caiming Xiong

Paper link: https://arxiv.org/pdf/2110.08222

Abstract

To study the problem of Fact-Checking in Dialogue, we construct and introduce DIALFACT, a testing benchmark dataset crowd-annotated conversational claims, paired with pieces of evidence from Wikipedia. There are three sub-tasks in DIALFACT: 1) Verifiable claim detection task distinguishes whether a response carries verifiable factual information; 2) Evidence retrieval task retrieves the most relevant Wikipedia snippets as evidence; 3) Claim verification task predicts a dialogue response to be supported, refuted, or not enough information.


Dataset Details

The statistics for the Test and Validation sets are shown in the figure below.

Data stats

The test and validation are present in the data folder of this repo.

The synthetic training dataset for AugWoW model is present here: data

Data format

Description of keys and values present in the dataset files:

{
    "context_id": "Context ID",
    "id": "Context ID --- ResponseID",
    "data_type": "Type of response: generated or written",
    "context":"List of utterances in dialogue history",
    "response": "The claim or response",
    "evidence_list": "List of evidences. Eack item in list is a list of following:"
        ["Wikipedia page Title","Wikipedia Link","Test snippet shown.","an index - not useful for the task", "optionally present value gt_evidence_added - indicates an evidence which belonged to the original utterance in WoW added for NEI claims." ],
    "response_label": "One of the three labels: SUPPORTS, REFUTES, NOT ENOUGH INFO",
    "type_label": "If the response is factual (Verifiable) or personal (Non-Verifiable)"
}

Results

The results for claim verification on test set. Test Results

The results for claim verification on validation set. Validation Results


Citation

@article{gupta2021dialfact,
  title={DialFact: A Benchmark for Fact-Checking in Dialogue},
  author={Gupta, Prakhar and Wu, Chien-Sheng and Liu, Wenhao and Xiong, Caiming},
  journal={arXiv preprint arXiv:2110.08222},
  year={2021}
}

Questions?

For any questions, feel free to open issues, or shoot emails to

License

The code is released under BSD 3-Clause - see LICENSE for details.

About

We construct and introduce DIALFACT, a testing benchmark dataset crowd-annotated conversational claims, paired with pieces of evidence from Wikipedia.

Topics

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages