Research map / Vision-language and multimodal / Cross-modal learning
Enhancing Meme Emotion Understanding with Multi-Level Modality Enhancement and Dual-Stage Modal Fusion
Yi Shi, Wenlong Meng, Zhenyuan Guo, Chengkun Wei, Wenzhi Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace dual-stage fusion with simple concatenation (ablation) | MemoDetector · MET-MEME | Macro-F1 (%): 43.57 -> 45.33 | Improved |
| add Text Meaning enhancement step (ablation) | MemoDetector · MOOD | Macro-F1 (%): 80.54 -> 82.65 | Improved |
| add Image Description step (ablation) | MemoDetector · MET-MEME | Macro-F1 (%): 44.29 -> 45.33 | Improved |
| add Combined Implicit Meaning step (ablation) | MemoDetector · MET-MEME | Macro-F1 (%): 44.32 -> 45.33 | Improved |
| add Context Analysis step (ablation) | MemoDetector · MOOD | Macro-F1 (%): 80.75 -> 82.65 | Improved |
| use four-step enhancement instead of direct single-step enhancement (ablation) | MemoDetector · MET-MEME and MOOD | Accuracy, Precision, Recall, F1 | Improved |
| use bidirectional cross-attention in second fusion stage instead of add, concat or cross-attention (ablation) | MemoDetector · MET-MEME and MOOD | Accuracy, Macro-F1 | Improved |
| add chain-of-thought prompting to MLLM direct classification | Qwen2.5-VL-7B · MET-MEME | Macro-F1 (%): 32.30 -> 36.59 | Improved |
| add chain-of-thought prompting to MLLM direct classification | Qwen2.5-VL-32B · MET-MEME | Macro-F1 (%): 35.30 -> 32.74 | Hurt |
| use 32B MLLM rather than 7B for text enhancement (ablation) | MemoDetector · MET-MEME | Macro-F1 (%): 41.22 -> 45.33 | Improved |
| supervised fine-tune Qwen2.5-VL-7B | Qwen2.5-VL-7B · MET-MEME | Macro-F1 (%): 32.30 -> 41.03 | Improved |
| use MLLM-enhanced text with small model instead of MLLM zero-shot | Qwen2.5-VL-7B · MOOD | Macro-F1 (%): 48.04 -> 81.72 | Improved |
Closest papers by meaning
- Multi-Granular Multimodal Clue Fusion for Meme Understanding (AAAI 2025 · similarity 0.90)
- Uncertain Multimodal Intention and Emotion Understanding in the Wild (CVPR 2025 · similarity 0.87)
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.86)
- Hateful Meme Detection through Context-Sensitive Prompting and Fine-Grained Labeling (Student Abstract) (AAAI 2025 · similarity 0.86)
- Enriching Multimodal Sentiment Analysis Through Textual Emotional Descriptions of Visual-Audio Content (AAAI 2025 · similarity 0.86)
- CAMEL: Capturing Metaphorical Alignment with Context Disentangling for Multimodal Emotion Recognition (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
