Explainability of Text Clustering Visualizations - Twitter Disinformation Case Study

Abstract

While text clustering methods have been available for decades, there is a paucity of material that would help practitioners with the choice and configuration of suitable algorithms and visualizations. In this article, we present a case study analyzing two disinformation datasets composed of tweets from the era of the 2016 United States Presidential Election. We use this to demonstrate steps for selecting the best configuration of the clustering algorithm and consequently conduct a user experiment for evaluating the comprehensibility of three alternate visualizations. A supplementary GitHub repository contains source code with examples.