TrustedNews is a research project on trustworthy AI for the written press.
We are group of AI researchers and digital media practitioners working on the analysis of informational quality in press documents.
Objectives
Research questions: How objective or subjective is a press report? How vague or precise is it? Was the document AI generated? Which parts of a document are particularly checkworthy? Can the text be qualified as fake news and on what basis? Is the text driven by a specific agenda, such as propagandist motives, or is the intention to transparently inform?
Methods: We use neuro-symbolic methods to answer these questions, comparing corpora coming from different press groups and sources. We take advantage of the large data bank of the SIPA-Ouest France group, and rely on the expertise of journalists to define analytics with humans in the loop.
Applications: The project is intended for expert as well as lay users. It aims to leverage our analytics for media literacy and content enhancement.
News
Keep posted, we will seek to recruit an AI engineer or postdoc.
Consortium
- IRL Crossing, CNRS
- IRISA, CNRS, LinkMedia team
- LIP6
- Airbus
- Ouest-France
Context
The project is an Astrid Maturation project (2026-2028), funded by the Agence Innovation Defense and operated by the Agence Nationale de la Recherche. The project is an elaboration of the earlier project Hybrinfox (2022-2024, Institut Nicod/IRISA/Mondeca/Airbus).
Publications
« Measuring Vagueness-Based Persuasion Techniques in an AI-Generated Influence Campaign« , Infox sur Seine 2026.
« Embedding Style Beyond Topics: Analyzing Dispersion Effects Across Different Language Models » COLING 2025.
« HYBRINFOX at CheckThat! 2024 – Task 1: Enhancing Language Models with Structured Information for Check-Worthiness Estimation » CLEF 2024, CheckThat! Lab
« HYBRINFOX at CheckThat! 2024 – Task 2: Enriching BERT Models with the Expert System VAGO for Subjectivity Detection » CLEF 2024, CheckThat! Lab (Winner method for English, 1st of 15)
« Exposing Propaganda: an Analysis of Stylistic Cues Comparing Human annotations and machine classification » EACL 2024, Third Workshop on Understanding Implicit and Underspecified Language
« A Multilabel Dataset of French Fake News: Human and Machine Insights » LREC-COLING 2024
« Measuring Vagueness and Subjectivity in Texts: from symbolic to neural VAGO » 2023 IEEE International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT).
« Un traitement hybride du vague textuel: du système expert VAGO à son clone neuronal » TALN 2023.
« VAGO: un outil en ligne de mesure du vague et de la subjectivité » Conférence nationale sur les applications pratiques de l’intelligence artificielle (APIA 2002).
« Combining Vagueness Detection with Deep-Learning to Identify Fake News » FUSION 2021, 24th international conference on Information Fusion.
Contact
Paul Égré, IRL Crossing CNRS, paul.egre@cnrs.fr
