claimreview-data

This repository contains a dataset of claims and their corresponding fact-checks.

The data is automatically updated every day with ClaimReview.

You can download the latest version from the release section, or browse all the versions.

Collection

The data collection is performed in 6 steps:

1. Collection of ClaimReviews URLs Candidates: using DataCommons and Google Fact-Check API, we get all the URLs where ClaimReviews are published
2. Collection of ClaimReviews from fact-checkers: we recollect from the websites of fact-checkers the ClaimReviews
3. Validation and Cleaning: we fix and clean the metadata
4. Ratings Mapping: we normalise the labels to credible, mostly credible, uncertain, mostly non-credible, non-credible and unknown
5. Occurrences Extraction and Unshortening: we extract the URLs where the claims occur and we resolve the links that use shortening services (e.g., bit.ly) or archives (e.g., archive.is)
6. Misinformation Database and Snapshot: we build the output files described below

The process of collection is run by the claimreview_collector_full from the MisinfoMe project.

Output files

Each archive contains the following files:

• ifcn_sources.json: details of the fact-checkers present in the dataset, such as website, country and language. The details also include the details of the IFCN compliance: date of issue, expiration, adherence to each of the skills (e.g. transparency of sourcing, transparency of methodology).
• claim_reviews_raw.json: this file contains the ClaimReviews collected in the first step from DataCommons and Google Fact-Check Tools. As noted before, this is bigger than the final cleaned dataset (recollection issues) but contains uncleaned data (appearance and firstAppearance fields).
• claim_reviews_recollected.json: the recollected dataset.
• claim_reviews.json: the final cleaned dataset with ratings mapping and unshortening.
• claim_labels_mapping.json: statistics on how labels have been translated.
• disagreeing_reviews.json: cases where the same URL has received multiple disagreeing ratings.
• not_ifcn_sources.json: this file contains a list of domains that published ClaimReview but are not in the IFCN list.
• links_all_full.json: details of the reviewed URLs. For each URL that has been reviewed, we show the reviews and the normalised ratings.
• stats.json: statistics of data collection.

This data is currently used by MisinfoMe.

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.