_alexcristea@EEUCA 2026: A Robust Early-Fusion ERNIE Pipeline for Multimodal COVID-19 Vaccine Meme Classification
Year: 2026
Type: conference-paper
Abstract: This paper presents our team's submission for the EEUCA 2026 shared task on Multimodal Vaccine Critical Meme Detection.To tackle the inherent challenges of internet sarcasm and high label noise, we propose a robust, heavily regularized early-fusion text pipeline.Bypassing computationally heavy visual encoders, we extract text directly from meme images via OCR, concatenate it with the user's social media post, and process the unified context through an ERNIE-2.0-Largeencoder.To combat severe overfitting, we replace the standard classification head with a Multi-Sample Dropout architecture (p = 0.3).Our optimized, lightweight text-only pipeline achieved a peak Macro F1 score of 0.834, securing 4th place on the official leaderboard.Furthermore, an ablation study utilizing Focal Loss demonstrates that our primary solution using standard Cross-Entropy provides superior robustness against the inherent label noise found in internet meme datasets.
Authors: Cristea Alexandru-Marian, Costin Ionescu
Institution: University of Bucharest
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7166884511