A Coefficient of Determination for Topic Models

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Thomas Jones*
• Companies: 3e Services LLC
• Keywords: Topic Modeling ; Text Mining ; Multinomial Distribution ; R-squared ; Coefficient of Determination ; Latent Dirichlet Allocation

Abstract

This document proposes a new (old) metric for evaluating goodness of fit in topic models, the coefficient of determination, or R-squared. Within the context of topic modeling, R-squared has the same interpretation that it does when used in a broader class of statistical models. The topic model R-squared uses a geometric interpretation of the standard R-squared statistic. Reporting R-squared with topic models addresses two current problems in topic modeling: a lack of standard evaluation metrics and ease of communication with lay audiences.