Estimating and Visualizing the Time-varying Effects of a Binary Covariate on Longitudinal Big Text Data

Abstract

We propose a method to estimate and visualize effects of a binary covariate on the longitudinally observed text data. Our method consists of series of analytical methods: extracting keywords through a morphological analysis, estimating the time-varying regression coefficient of a binary covariate for keyword's appearance and frequency, classifying summary of estimates, and visualizing the time-varying effects of a binary covariate in animated scatter plots. The procedure is demonstrated with Peace Declaration text data observed for forty years in two cities.

Keywords

• Text mining
• Summarization
• Semiparametric modeling
• Classification
• Animation
• Knowledge discovery.