Data will be shared exclusively for the purpose of participating in the shared task CheckThat!Lab 2026 Task 1 - Source Retrieval for Scientific Web Claims.

To use the dataset, you need to agree to the following Declaration of Commitment:

DECLARATION OF COMMITMENT I hereby make a legally binding declaration regarding the use of this dataset provided by the organizers of the CheckThat!Lab 2026 Task 1:
1. I am aware that the corpora made available may, in their entirety or in individual parts, be protected by copyright, are subject to strict licensing conditions, and may contain potentially personal data. I undertake to use the corpora exclusively for project-related research within the above mentioned context, not to use the corpora, in whole or in part, outside this specific research context, and in particular not to reproduce them for any other purpose or make them accessible to third parties for other purposes. After completion of the research in this context, and at the latest upon completion of the project, I will delete the data. I will not circumvent the technical and organizational measures implemented to protect the data.

2. I am aware that situations may arise in which research results from the project may not, or not immediately, be disseminated or made publicly accessible for legal reasons. Even in cases of doubt, I will comply with a legally justified assessment by the CheckThat! Lab 2026 Task 1 team.

3. I declare that, in the event of at least negligent violation of the aforementioned obligations, I will indemnify the CheckThat! 2026 Task 1 team against any resulting claims for damages by third parties.

Please be sure to provide your full legal name, address, and affiliation name.

Overview

This repo contains the data for CheckThat!Lab 2026 - Task 1: Source Retrieval for Scientific Web Claims

Task Definition

Source Retrieval for Scientific Web Claims: Given a social media post that contains a scientific claim and an implicit reference to a scientific paper (mentions it without a URL), retrieve the mentioned paper from a pool of candidate papers.

Datasets

The collection set contains information about 10k publications. The query sets (train and dev) comprise English, German, and French social media posts.

collection_data.json collection set with 10k publications (columns: pubkey, title, abstract, venue / location, authors)
de_train.json DE training data (columns: index, pubkey, text)
de_dev.json DE development data (columns: index, pubkey, text)
en_train.json EN training data (columns: index, pubkey, text)
en_dev.json EN development data (columns: index, pubkey, text)
fr_train.json FR training data (columns: index, pubkey, text)
fr_dev.json FR development data (columns: index, pubkey, text)

Usage

Below is an example for usage:

lang = "fr"
split = "train"

data = load_dataset("sschellhammer/CT26_Task1_SourceRetrievalForScientificWebClaims", lang)
split_data = data[split]
labels = np.array(split_data["pubkey"])

Make sure your runtime is authenticated with an account that was granted access to this dataset.

Downloads last month: 44