Explainable Topic Extraction with Decision Trees

#

Purpose

#

This project is a collection of a couple different experiments and utilities for topic extraction, including:

• Comparing popular topic extraction libraries on different kinds of documents
• A simple implementation of decision trees to explain group membership for clustered texts
• Some additional utilites for explaining decision tree rules based on bag of word embedding features
• Plotting utilities for derived clusters and/or their assigned topics

Datasets

#

• 20 Newsgroups
• Bitext Customer Support
• ShopperSentiments
• Social Media Sentiments Analysis Dataset

Topic Extractors

#

• yake
• pytextrank
• lda
• nmf
• word count
• tfidf score
• decision tree

Outputs

#

• The results report outlining approach and findings
• The outputs folder, which contains breakdowns of dataset topics, topic evaluations, and tests of group differences
• Decision tree topic extraction implementation