Automated Insights on Visualizations with Natural Language Generation

Abstract

Quantitative data, such as a 10k financial report, requires cognitive effort to scan the columns and rows and identify patterns and important takeaways, whether novice or subject matter expert. Visualizations can be used to summarize and reveal patterns. However, unless a visualization contains arrows or other callouts, it still requires cognitive effort to understand and rank the important conclusions to which a reader should pay attention. In this research, we aim to reduce the cognitive effort in understanding tabular data by combining charts with ranked natural language generated (NLG) bullet point statements that summarize the top takeaways. The contribution of this work is an NLG pipeline to computationally extract insights from tabular data and provide textual comments, which are then integrated with visualizations of the same data set.