Fake News Detection via Intermediate-Layer Emotional Representations
Résumé
This paper introduces Intermediate-Layer Emotional Representations (ILER), a novel emotional embedding technique designed to capture nuanced emotional patterns in textual data. While we demonstrate its utility in the domain of fake news detection, ILER was conceived as a generalizable framework to be applied across various tasks requiring emotion or sentiment analysis.
We aim to leverage the abstract and rich representations offered by ILER by combining publisher-driven emotions embedded within news content and social emotions derived from user comments. This approach seeks to capture the intricate emotional nuances often associated with fake news dissemination. Evaluations on benchmark datasets emulating live systems show a 5-point increase in F1-score and a 3.9-point increase in accuracy compared to state-of-the-art emotion representations, including Dual Emotion Features. Our findings underscore the potential of abstracted emotional representations to deepen context understanding, enabling more robust and precise emotionally assisted fake news detection.
Mots clés
- Intermediate Layer Representations
- Social Media
- CCS Concepts Information systems → Information retrieval • Computing methodologies → Natural language processing Neural networks Fake News Detection Emotion Emotion Representation Intermediate Layer Representations Social Media
- CCS Concepts
- Information systems → Information retrieval
- • Computing methodologies → Natural language processing
- Neural networks Fake News Detection
- Emotion
- Emotion Representation
Domaines
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |