Conduct various tests for evaluating implicit biases in word embeddings: Word Embedding Association Test (Caliskan et al., 2017), <doi:10.1126/science.aal4230>, Relative Norm Distance (Garg et al., 2018), <doi:10.1073/pnas.1720347115>, Mean Average Cosine Similarity (Mazini et al., 2019) <arXiv:1904.04047>, SemAxis (An et al., 2018) <arXiv:1806.05521>, Relative Negative Sentiment Bias (Sweeney & Najafian, 2019) <doi:10.18653/v1/P19-1162>, and Embedding Coherence Test (Dev & Phillips, 2019) <arXiv:1901.07656>.
- Semantics derived automatically from language corpora contain human-like biases
- Word embeddings quantify 100 years of gender and ethnic stereotypes
- A Transparent Framework for Evaluating Unintended Demographic Bias in Word Embeddings
- Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings
- SemAxis: A Lightweight Framework to Characterize Domain-Specific Word Semantics Beyond Sentiment
- Attenuating Bias in Word Vectors
This page was built for software: sweater
