CMU-MOSEI, the full name of which is CMU Multimodal Opinion Sentiment and Emotion Intensity, is the largest trimodal dataset with two types of annotations: sentiment and emotion.

The CMU-MOSEI dataset contains more than 65 hours of annotated videos from more than 1,000 speakers covering 250 topics. It is worth noting that the dataset has a multi-label feature, that is, each sample may correspond to more than one emotion, and the strength of the corresponding emotion is also different, ranging from [-3 to 3].