MissBench
收藏资源简介:
MissBench是由越南国立大学团队构建的多模态情感分析基准,整合了CMU-MOSI、CMU-MOSEI、IEMOCAP和CH-SIMS四个经典数据集,涵盖英语和中文的情感识别与情绪分析任务。该数据集创新性地设计了共享缺失率(SMR)和不平衡缺失率(IMR)两种协议,通过标准化数据分割和掩码种子实现可复现评估。其核心价值在于提出了模态公平指数(MEI)和模态学习指数(MLI)两大诊断指标,能有效量化不同模态在缺失条件下的贡献均衡性与优化动态,为现实场景中不完整多模态数据的研究提供系统性评估框架。
MissBench is a multimodal sentiment analysis benchmark constructed by the team from Vietnam National University. It integrates four classic datasets, namely CMU-MOSI, CMU-MOSEI, IEMOCAP and CH-SIMS, covering sentiment recognition and emotion analysis tasks in both English and Chinese. This dataset innovatively designs two protocols: Shared Missing Rate (SMR) and Imbalanced Missing Rate (IMR), and enables reproducible evaluation through standardized data splits and masking seeds. Its core value lies in proposing two diagnostic indicators, Modal Fairness Index (MEI) and Modal Learning Index (MLI), which can effectively quantify the contribution balance and optimization dynamics of different modalities under missing conditions, providing a systematic evaluation framework for research on incomplete multimodal data in real-world scenarios.




