Clarification on Annotation Order for CH-SIMS Dataset

[ilucasgoncalves]

Hello!

Thank you for sharing the CH-SIMS dataset—it’s a valuable resource for multimodal sentiment analysis, and I'm excited to work with it!

I’m looking for clarification regarding the order of the modality-specific annotations in the provided CSV file. Here’s an example row from the file for reference:

video_0001  1  [transcription text]  -1  -1  -1   -1  Negative  train
video_0001  2  [transcription text]   1   1  0.8   1  Positive  train

In this row, there are four numerical sentiment labels following the [transcription text]. Based on the dataset description in your paper, I understand these labels likely correspond to the Text-only, Audio-only, Visual-only, and Multimodal annotations. Could you please confirm the exact order of these columns?

Specifically, is the order of the labels? See below:

1. Text-only
2. Audio-only
3. Visual-only
4. Multimodal

Or is it arranged differently?