Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Sep 4·5 min read
WELD: The First Naturalistic Long-Period Small-Team Workplace Emotion Dataset for Ubiquitous Affective Computing
TL;DR
WELD is the first extensive dataset capturing workplace emotions over 30 months, providing insights into emotional dynamics within small teams. It reveals significant findings on emotional variance and biases in facial expression recognition models.
✦ Why It Matters
Utilize the WELD dataset to enhance emotion recognition models and address biases in AI applications today.
Key Takeaways
- Explore WELD for insights into workplace emotional dynamics.
- Use findings to improve predictive models for emotional states.
- Address biases in FER models using WELD's data.
- Leverage longitudinal data for team-based emotional analysis.
Full Summary
WELD (Workplace Emotion Longitudinal Dataset) is a groundbreaking dataset that spans over 30 months, capturing emotional expressions of 49 employees in a Chinese software company. It includes 733,780 facial-expression probability vectors categorized into seven classes, enabling both longitudinal studies of individual emotions and relational analyses within teams.
The dataset was validated by replicating known emotional phenomena and uncovering new insights, such as the influence of seasonality on emotional variance and the identification of six distinct emotional states. Notably, it highlights a concerning bias in facial expression recognition (FER) models, which over-predict anger in neutral Asian faces compared to Western norms.
This dataset is crucial for advancing affective computing and ensuring fairness in emotion recognition technologies.