SemMSA-data

Processed features of CMU-MOSI, CMU-MOSEI and CH-SIMS used by SemMSA: Latent Semantic-Aided Robust Multimodal Sentiment Analysis with Incomplete Data (paper, code, project page).

These are the processed (unaligned) features released with MMSA, which are also used by LNLN and P-RMF.

File	Dataset	Speakers	Clips	Train	Valid	Test	Language	Size
mosi_unaligned_50.pkl	CMU-MOSI	93	2,199	1,284	229	686	English	0.6 GB
mosei_unaligned_50.pkl	CMU-MOSEI	1,000	22,856	16,326	1,871	4,659	English	13.7 GB
sims_unaligned_39.pkl	CH-SIMS	474	2,281	1,368	456	457	Chinese	1.2 GB

Usage

git clone https://github.com/peacelwh/SemMSA.git && cd SemMSA
bash scripts/download_data.sh          # all three datasets -> data/

or directly:

from huggingface_hub import hf_hub_download
path = hf_hub_download("Bluexxx/SemMSA-data", "sims_unaligned_39.pkl", repo_type="dataset")

mosei_unaligned_50.pkl (13.7 GB) is stored as 13 parts in mosei_unaligned_50/. download_data.sh reassembles it; to do it by hand:

huggingface-cli download Bluexxx/SemMSA-data --repo-type dataset --include "mosei_unaligned_50/*" --local-dir .
cat mosei_unaligned_50/mosei_unaligned_50.pkl.part* > mosei_unaligned_50.pkl
sha256sum mosei_unaligned_50.pkl   # ad8b23d50557045e7d47959ce6c5b955d8d983f2979c7d9b7b9226f6dd6fec1f

Format

Each pickle is a dictionary with train / valid / test splits. Every split contains:

Key	Content
text_bert	BERT token ids, attention mask and segment ids, (N, 3, L)
vision	visual features, (N, T_v, d_v)
audio	acoustic features, (N, T_a, d_a)
audio_lengths, vision_lengths	valid sequence lengths
regression_labels	sentiment scores, in [-3, 3] for MOSI / MOSEI and [-1, 1] for SIMS
raw_text, id	transcript and clip id

License and citation

The features are derived from the original datasets and are redistributed for non-commercial research purposes only. They remain subject to the licenses and terms of their original releases. If you use them, please cite the original datasets:

@article{zadeh2016mosi,
  title   = {Multimodal Sentiment Intensity Analysis in Videos: Facial Gestures and Verbal Messages},
  author  = {Zadeh, Amir and Zellers, Rowan and Pincus, Eli and Morency, Louis-Philippe},
  journal = {IEEE Intelligent Systems},
  year    = {2016}
}

@inproceedings{zadeh2018mosei,
  title     = {Multimodal Language Analysis in the Wild: {CMU-MOSEI} Dataset and Interpretable Dynamic Fusion Graph},
  author    = {Zadeh, AmirAli Bagher and Liang, Paul Pu and Poria, Soujanya and Cambria, Erik and Morency, Louis-Philippe},
  booktitle = {Proceedings of ACL},
  year      = {2018}
}

@inproceedings{yu2020sims,
  title     = {{CH-SIMS}: A {Chinese} Multimodal Sentiment Analysis Dataset with Fine-grained Annotation of Modality},
  author    = {Yu, Wenmeng and Xu, Hua and Meng, Fanyang and Zhu, Yilin and Ma, Yixiao and Wu, Jiele and Zou, Jiyun and Yang, Kaicheng},
  booktitle = {Proceedings of ACL},
  year      = {2020}
}

and SemMSA:

@article{li2026semmsa,
  title   = {SemMSA: Latent Semantic-Aided Robust Multimodal Sentiment Analysis with Incomplete Data},
  author  = {Li, Wenhao and Wu, Zhibin and Xiao, Chong and Wang, Qiangchang},
  journal = {arXiv preprint arXiv:2609.30238},
  year    = {2026}
}