LilyMeme@EEUCA 2026: Multimodal Vaccine Meme Stance Detection with Task-Adapted MemeCLIP and Complementary Ensembling

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

The LilyMeme@EEUCA 2026 approach addresses multimodal vaccine meme stance detection, a challenging task due to memes' implicit stances, sarcastic nuances, and complex cross-modal interactions. Developed for the VaxMeme Shared Task @EEUCA 2026, this system classifies vaccine-related memes into "Vaccine-critical," "Neutral," and "Pro-vaccine" categories. Building upon the MemeCLIP framework, the method incorporates several enhancements: task-specific adaptation, lightweight cross-modal fusion, noise-aware training, LLM-assisted semantic augmentation, and inference-stage optimization. The final predictions are generated by ensembling multiple complementary variants. This comprehensive approach achieved a Macro-F1 score of 0.8494 on the official test set, securing first place and highlighting the effectiveness of noise-aware training and late-stage ensembling for robust stance identification.

Key takeaway

For NLP Engineers building multimodal content moderation systems, this research demonstrates a winning strategy for complex meme analysis. You should integrate task-specific adaptations, noise-aware training, and LLM-assisted semantic augmentation into your MemeCLIP-based models. Employing late-stage ensembling of diverse model variants can significantly boost your system's Macro-F1 score, ensuring more robust and accurate detection of nuanced stances on sensitive topics like vaccination.

Key insights

Task-adapted MemeCLIP with noise-aware training and ensembling significantly improves vaccine meme stance detection.

Principles

• Memes require specialized multimodal analysis.
• Noise-aware training boosts robustness.
• Ensembling improves final prediction accuracy.

Method

The approach systematically enhances MemeCLIP via task-specific adaptation, lightweight cross-modal fusion, noise-aware training, LLM-assisted semantic augmentation, and inference-stage optimization, followed by complementary ensembling.

In practice

• Apply noise-aware training for noisy data.
• Use LLMs for semantic augmentation.
• Ensemble diverse models for robust results.

Topics

• Multimodal Stance Detection
• Vaccine Memes
• MemeCLIP
• Noise-Aware Training
• Ensemble Learning
• LLM Semantic Augmentation

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Adapting Reinforcement Learning with Chain-of-Thought Supervision for Explainable Detection of Hateful and Propagandistic Memes — Reinforcement learning with CoT supervision significantly improves MLLM performance and explainability for hateful meme detection.
• A Resource for Enthymeme Detection in Controversial Political Discourse — The study of annotator disagreement in enthymeme detection improves model performance and understanding of human inference.
• I built a public voting benchmark where models have to make memes out of daily news — Memebench evaluates LLM humor and creativity by generating memes from daily news for public A/B voting.
• Zero-shot Tweet-Level Stance Detection Enhanced by External Knowledge and Reflective Chain-of-Thought Reasoning — KIRP improves zero-shot tweet stance detection by integrating external knowledge, entity reorganization, and reflective Chain-of-Thought reasoning.
• Task-guided cross-subject latent alignment: a multi-encoder-decoder VAE — MED-VAE aligns neural activity across subjects using a pretrained ANN scaffold, eliminating the need for shared stimuli.
• I Know What You Meme, Even If it Emerged Today: Understanding Evolving Memes through Open-World Knowledge Acquisition — Meme understanding requires dynamic, open-world knowledge acquisition beyond static model parameters.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.