09. Text Bias Quiz

Bias Quiz

A recent study found that one popular model for encoding words and text, learns to represent problematic relationships between words, such as associating man:doctor and woman:nurse.

This model was given a large volume of training data: unlabeled text, taken from Google News. The text itself is reflective of Google users and journalists.

What kind of bias is responsible for the model learning these problematic word relationships?
SOLUTION: Data and model bias

Next Concept