Skip to content

jessijzhao/interpreting-word-vectors

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 

Repository files navigation

Interpretable Word Vector Subspaces

Presented at the ICLR 2020 workshop on Machine Learning in Real Life (ML-IRL) on April 26 2020. Link to paper here.

Abstract

Natural Language Processing relies on high-dimensional word vector representations, which may reflect biases in the training text corpus. Identifying these biases by finding corresponding interpretable subspaces is crucial for achieving fair decisions. Existing works have adopted Principal Component Analysis (PCA) to identify subspaces such as gender but fail to generalize efficiently or provide a principled methodology. We propose a framework for existing PCA methods and considerations for optimizing them. We also present a novel algorithm for finding topic subspaces more efficiently and compare it to an existing approach.

Contents

About

Identifying Interpretable Word Vector Subspaces with PCA

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published