Research map / Vision-language and multimodal / Human and embodied modeling
CMMA: Benchmarking Multi-Affection Detection in Chinese Multi-Modal Conversations
Yazhou Zhang, Yang Yu, Qing Guo, Benyou Wang, Dongming Zhao, Sagar Uprety, Dawei Song, Qiuchi Li, Jing Qin
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 |
|---|---|---|---|
| add video (ResNet) and audio (VGGish) to BERT text-only model (ablation) | BERT multimodal baseline · CMMA sentiment | macro-F1: 54.89 -> 73.29 | Improved |
| replace EfficientNet video encoder with ResNet in BERT trimodal model (ablation) | BERT+VGGish model · CMMA sentiment | macro-F1: 71.12 -> 73.29 | Improved |
| replace BERT text encoder with GPT-2 in trimodal model (ablation) | ResNet+VGGish model · CMMA sarcasm | macro-F1: 75.23 -> 70.67 | Hurt |
| replace BiLSTM with BERT as text encoder (ablation) | text-only model · CMMA sentiment | macro-F1: 50.74 -> 54.89 | Improved |
| add audio to BERT+ResNet bimodal model (ablation) | BERT+ResNet · CMMA sarcasm | macro-F1: 66.75 -> 75.23 | Improved |
| use multi-head attention instead of multiplication for fusion (ablation) | trimodal model · CMMA sentiment | accuracy: 72.22 -> 78.48 | Improved |
| use multi-head attention instead of concatenation for fusion (ablation) | trimodal model · CMMA emotion | accuracy: 76.32 -> 77.09 | Improved |
| increase conversational context length from 0 to 2 (ablation) | trimodal model · CMMA sentiment | macro-F1: 73.29 -> 76.91 | Improved |
| increase conversational context length from 0 to 1 (ablation) | trimodal model · CMMA sentiment | macro-F1: 73.29 -> 71.36 | Hurt |
| increase conversational context length from 0 to 2 (ablation) | trimodal model · CMMA emotion | macro-F1: 61.76 -> 68.79 | Improved |
| add speaker information (ablation) | trimodal model · CMMA humor | macro-F1: 75.91 -> 72.06 | Hurt |
| add speaker information (ablation) | trimodal model · CMMA emotion | macro-F1: 61.76 -> 64.23 | Improved |
Closest papers by meaning
- Adaptive Graph Learning for Multimodal Conversational Emotion Detection (AAAI 2024 · similarity 0.86)
- Can Large Language Models Help Multimodal Language Analysis? MMLA: A Comprehensive Benchmark (NeurIPS 2025 · similarity 0.86)
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.85)
- Friends-MMC: A Dataset for Multi-modal Multi-party Conversation Understanding (AAAI 2025 · similarity 0.85)
- A New Formula for Sticker Retrieval: Reply with Stickers in Multi-Modal and Multi-Session Conversation (AAAI 2025 · similarity 0.85)
- MIntRec2.0: A Large-scale Benchmark Dataset for Multimodal Intent Recognition and Out-of-scope Detection in Conversations (ICLR 2024 · similarity 0.85)
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.
