Claim2Source at CheckThat! 2026 Improving Multilingual Scientific Claim–Source Retrieval with Verification-based Re-Ranking

Jul 16, 2026

Upload by Hazel

Multilingual scientific claim–source retrieval identifies the publication supporting a scientific claim shared on social media. The task is difficult because claims may differ from source papers in language, wording, and granularity, weakening semantic links to evidence. A multi-stage retrieval framework for the CheckThat! 2026 Lab Task 1 combines structured claim and source representations, progressive candidate refinement, and language-specific dense retrieval adaptation. Similarity-based re-ranking and verification-based re-ranking select the best supporting source using verification signals. The method reaches an average MRR@5 of 0.7628 on English, German, and French claims and ranks first on the leaderboard.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

Introduction

Related Work

Claim2Source at CheckThat! 2026: Improving Multilingual Scientific Claim–Source Retrieval with Verification-based Re-Ranking Notebook for the CheckThat! Lab at CLEF 2026 Tobias Schreieder 1, * , Harsh Khandelwal2, Yu-Ling Zhong1 and Michael Färber1 1 TU Dresden & ScaDS. AI Dresden/Leipzig, Dresden, Germany 2Friedrich-Alexander University of Erlangen–Nuremberg, Erlangen, Germany Abstract Multilingual scientific claim–source retrieval aims to identify the scientific publication supporting a claim shared on social media. This task is challenging because claims often differ from source publications in terms of language, wording, and level of detail, which weakens the connection between claims and their underlying evidence. In this paper, we present our approach for the CheckThat! 2026 Lab Task 1: Source Retrieval for Scientific Web Claims. We propose a multi-stage retrieval framework for multilingual scientific claim–source retrieval that combines structured claim and source representations with progressive candidate refinement. To address multilingual retrieval challenges, the framework employs bilingual claim representations, metadata-enhanced source representations, and language-specific adaptation of dense retrieval models. Building on this setup, a first-stage retriever generates an initial pool of candidate sources, after which similarity-based re-ranking improves the ranking of highly relevant sources and verification-based re-ranking identifies the candidate source that best supports the claim using verification signals. Our approach achieves an average MRR@5 score of 0.7628 across English, German, and French claims, ranking first on the CheckThat! 2026 leaderboard. Keywords Fact Verification, Scientific Claim-Source Retrieval, Multilinguality, Re-Ranking, Large Language Model 1. Introduction Social media platforms such as X (formerly Twitter) have become important channels for communicating and discussing scientific findings, particularly in fast-moving domains such as health and biomedicine [1] . Scientific claims shared online often discuss research findings without explicitly attributing them to their original scientific source. Identifying this source is challenging because claims often contain only partial descriptions of research findings and may rephrase or simplify the original evidence, creating a semantic mismatch between claims and their underlying publications [2, 3] . This problem becomeseven more difficult when claims and source documents are written in different languages, requiring retrieval systems to align semantically equivalent content across heterogeneous representations [4, 5] . Attributing claims to their supporting evidence is an important component of automated fact verification systems, which aim to distinguish factual information from misinformation shared online. This problem is studied in the CheckThat! 2026 Lab, which focuses on advancing multilingual factchecking [6], through Task 1: Source Retrieval for Scientific Web Claims [7] . The task was introduced in 2025 [8] and is extended in 2026 toward a multilingual setting with English, German, and French claims [7] . Throughout this paper, we refer to this task as multilingual scientific claim–source retrieval. While previous systems mainly focused on improving retrieval performance through multi-stage retrieval and re-ranking pipelines [9, 10], cross-lingual retrieval introduces additional challenges, as CLEF 2026 Working Notes, 21 – 24 September 2026, Jena, Germany * Corresponding author. $ tobias.schreieder@tu-dresden.de (T. Schreieder); harsh.khandelwal@fau.de (H. Khandelwal); yu-ling.zhong@mailbox.tu-dresden.de (Y. Zhong); michael.faerber@tu-dresden.de (M. Färber) 􀀚 0009-0000-8268-4204 (T. Schreieder); 0009-0004-0237-5735 (H. Khandelwal); 0009-0007-2361-4944 (Y. Zhong); 0000-0001-5458-8645 (M. Färber) © 2026 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution

FAQs

What is the goal of multilingual scientific claim–source retrieval in CheckThat! 2026?

It aims to identify the scientific publication that supports a claim shared on social media, even when the claim and the source are written differently.

Why is claim–source linking harder in a multilingual setting?

Claims may rephrase or simplify evidence and, in multilingual cases, the claim and source documents use different languages, creating semantic and representation mismatches.

How does the proposed system improve ranking beyond similarity-based retrieval?

It uses a three-stage pipeline: dense candidate generation, similarity-based re-ranking, and a final verification-based re-ranking stage that incorporates verification signals with an LLM to select the best supporting source.

Jul 16, 2026

Upload by Hazel