The ISEAR data was collected through questionnaires administered to university students in 37 countries between the 1980s and the early 1990s by Klaus R. Scherer (University of Geneva) and Harald G. Wallbott (University of Salzburg). Respondents were asked to recall experiences of seven fundamental emotions and answer detailed questions about their reactions and appraisals.
Classification
We released our version of the dataset and trained an emotion classifier to support use cases that involve research or the creation of more human AI systems.
These use cases may include emotional expression research, healthcare applications for mental health support, and technology-supported team coaching programs.
The classifier's input is the natural language description of a situation (SIT) and the output is the classification into one of seven labels:
| Label | Emotion | Definition |
|---|---|---|
| 0 | Joy | Happiness, pleasure, or satisfaction |
| 1 | Fear | Anxiety about threat or danger |
| 2 | Anger | Strong displeasure or antagonism |
| 3 | Sadness | Feeling of loss or disadvantage |
| 4 | Disgust | Aversion or revulsion |
| 5 | Shame | Embarrassment or humiliation |
| 6 | Guilt | Feeling responsible for wrongdoing |
The model is best suited for scenarios where an approximation of human emotions is desirable based on a situation description.
The dataset
The ISEAR (International Survey on Emotion Antecedents and Reactions) dataset contains 7,666 personal accounts of emotional experiences collected by psychologists Klaus R. Scherer and Harald G. Wallbott as part of a large-scale international study of emotions. Respondents from 37 countries were asked to describe situations where they experienced specific emotions (joy, fear, anger, sadness, disgust, shame, or guilt), along with detailed information about their reactions and appraisals.
Our dataset is a cleaned version of the original ISEAR questionnaire data, prepared for natural language processing and emotion classification tasks.
Ethical considerations and restrictions
The dataset and the model are released under Apache License 2.0.
We published this dataset and model with appropriate ethical guidelines that are available on the dataset card and model card, respectively.
This dataset should not be used for commercial emotion categorization systems to infer the emotions of a natural person.
Model availability
Dataset link on HuggingFace: savalera/isear-from-original
Model link on HuggingFace: savalera/isear-emotion-classifier
References
Official dataset source: Geneva Emotion Research Group
Scherer, K. R., & Wallbott, H. G. (1994). Evidence for universality and cultural variation of differential emotion response patterning. Journal of Personality and Social Psychology, 66(2), 310–328.