Open a1da4 opened 1 year ago
They analyse the intersectional biases (e.g. non-white & female) over time.
Embedding-based: they used WEAT metric on trained word2vec models.
Lexicon-based: they used the NRC-VAD lexicon. In this model, each usage is associated with three labels (dominance, valence, and arousal).
From this figure, females and Caribbean countries are more attributed to "family".
From this figure, non-white males and females achieve the highest levels of associations.
1. What is it?
2. What is amazing compared to previous works?
They analyse the intersectional biases (e.g. non-white & female) over time.
3. Where is the key to technologies and techniques?
Embedding-based: they used WEAT metric on trained word2vec models.
Lexicon-based: they used the NRC-VAD lexicon. In this model, each usage is associated with three labels (dominance, valence, and arousal).
4. How did evaluate it?
5. Is there a discussion?
6. Which paper should read next?