2026-10-14T01:13:33Z https://www.scidb.cn/oai

10.57760/sciencedb.30368 2025-10-24T09:36:14Z

2025-10-24 Multimodal Sentiment Analysis Experimental Data (Processed MOSI/MOSEI Features & Splits) doi:10.57760/sciencedb.30368 en [Overview]This release targets reproducible experiments for multimodal sentiment/emotion analysis. It provides de-identified, aligned features and splits derived from the public CMU-MOSI / CMU-MOSEI datasets. No raw audio/video is included—only processed text/audio/vision features and labels for quick reproduction and evaluation.[Structure]data/ - train.zip, valid.zip, test.zip (each contains: text.csv, audio.npz, vision.npz, labels.csv)code-snippets/ (loading + alignment/aggregation examples)docs/ (README, compliance statement)[Features] (edit to match your actual data)- Text: GloVe/BERT sentence-level vectors or word-aligned aggregation- Audio: COVAREP/opensmile stats (frame → segment aggregation)- Vision: OpenFace/Facet features (frame → segment aggregation)- Labels: sentiment intensity or emotion classes following the originals[Usage]Unzip data/* and use code-snippets/load_examples.py. Works on CPU or PyTorch-DirectML on Windows (tested on AMD 780M).[License & Notice]Released under CC BY-NC 4.0 (non-commercial, with attribution). Original data and terms belong to CMU-MOSI/MOSEI; this release contains derived features only. multimodal; sentiment analysis; emotion recognition; MOSI; MOSEI; feature alignment; reproducibility Chen Bo PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank