Description du projet
Une plateforme innovante pour démystifier les contenus de désinformation
La désinformation constitue une menace importante pour les démocraties modernes. De nouvelles solutions sont donc nécessaires pour en démystifier le contenu avec succès. Le projet AI4Debunk, financé par l’UE, vise à créer quatre interfaces alimentées par l’IA et centrées sur les humains. Ces interfaces comprennent un module web, une plateforme collaborative, une application pour smartphone et une interface de réalité augmentée (RA), qui seront toutes construites à l’aide d’une interface de programmation d’applications (API) de démystification. Le projet se concentrera sur la désinformation concernant la guerre en Ukraine et le changement climatique, et cartographiera le contenu de la désinformation sur des graphes de connaissances et le comparera aux bases de données existantes. La plateforme, baptisée Disinfopedia, est conçue pour détecter instantanément les faux contenus. L’application vérifie le contenu dans la vie de tous les jours, et l’interface de RA donne un aperçu des fausses informations et des réseaux sociaux.
Objectif
"AI4Deunk aims to develop 4 human-centered AI-powered interfaces: a web plug-in, a collaborative platform, a smartphone app, and an AR interface. All these solutions will be built upon a ""debunking"" API.
To develop such an API, AI4Debunk will focus on 2 peculiar disinformation topics: the war in Ukraine and disinformation related to climate change. By deeply analysing these case studies thanks to sociological experts (targets groups, sources, way of diffusion...) we will map disinformation contents on knowledge graphs, and compare them to existing databases thanks to AI methods such as NLP. These knowledge graphs will be the skeleton of our technical work: one will be multimodal meaning that all types of modalities (written content, image, video) will be directly embedded, while the monomodal will be only about written content (the different modalities will be previously embedded into written content).
Then debunking modules will be developed, for each type of modality, based on AI and ML methods that could detect and examine questionable content, by comparing it to the context extracted from the knowledge graphs. RNN, CNN, multi-modal language interpretation, and transformers methods will be used to that extent. The goal of such modules is to give a score of fakeness, which we will call Disinofscore, for content given as input.
The API will then integrate this method and will be the bases of the 4 interfaces abovementioned. The plugin will allow people to be instantly noticed when a content they are considering is fake or true, the collaborative platform - also called Disinfopedia, will allow more proactive online users to report questionable content, that will be checked by a debunking committee chaired by a senior analyst, the app interface will allow people to check content in their everyday life, while the AR interface will allow citizens to have hindsight on the incoming generation of fake news and social media."
Champ scientifique
- natural sciencescomputer and information sciencesdatabases
- natural sciencescomputer and information sciencesknowledge engineering
- engineering and technologyelectrical engineering, electronic engineering, information engineeringinformation engineeringtelecommunicationsmobile phones
- natural sciencesearth and related environmental sciencesatmospheric sciencesclimatologyclimatic changes
Mots‑clés
Programme(s)
Régime de financement
HORIZON-IA - HORIZON Innovation ActionsCoordinateur
1586 RIGA
Lettonie