- word embeddings provide a mathematical model for inferring meaning 
- previous research has shown that there tends to be gender/ethnic stereotype biases within these embeddings, especially when it comes to gender stereotypes around occupation

Why Analysis of Data Biases May Be An Issue

The algorithm in the NLP system may have it's own biases because it goes off a chosen model, training data, features, and word embeddings. 

Santa Clara University

1. Expression of Data
2. Analysis of Data
3. Interpretation of Data 

Potential Biases of Natural Language Processing

Straw, I and Callison-Burch, C. (2020). Research Article: Artificial Intelligence in mental health and the biases of language-based models. Plos One, Vol 15 (issue 12), 1-19. https://doi.org/10.1371/journal.pone.0240376

Research Article: Artificial Intelligence in mental health and the biases of language-based models

An individual will use their own language to express their emotions online as well as the data they produce.

Potential Biases of Natural Language Processing: Expression of Data

Potential Biases of Natural Language Processing: Analysis of Data

There may be biases in how practitioners interpret data and what they've already been trained to know. 

Learn Before

Related

Learn After