Interpreting Blackbox Text Classifiers with LDA-Based Topic Models

An LDA wrapper for explaining a blackbox classifier's predictions, as done in Oved, N., Feder, A. and Reichart, R. (2020) and presented in EMNLP 2020's blackbox workshop.

Currently supports only binary predictors.

The module was developed for domain-ruled data (see demo below), although data without domains is supported as well (see API below).

Installation

Git

Get the latest version using git (recommended): pip install git+https://github.com/EllRos/LDA-Explanation.git

Wheel

In order to avoid git, get the latest wheel build (might not be updated, but should be):

1. Download https://github.com/EllRos/LDA-Explanation/blob/main/dist/LDA_Explanation-0.0.1-py3-none-any.whl
2. run pip install LDA_Explanation-0.0.1-py3-none-any.whl from the download directory.

Note: While typically wanting to just run pip install https://github.com/EllRos/LDA-Explanation/blob/main/dist/LDA_Explanation-0.0.1-py3-none-any.whl, this might cause a strange BadZipFile error (even with pip cache disabled).

Requirements

Installation requiers (and includes) the installation of the following libraries (of any version):

• NumPy
• Pandas
• Matplotlib
• Gensim

Also requires Python version of 3.6 and above.

Documentation

API documentation: https://ellros.github.io/LDA-Explanation-docs/

Functionality and usage demonstration: https://ellros.github.io/LDA-Explanation/docs/demo/demo.html