Abstract This study builds on and responds to previous cultural analytics work on book reviews by comparing how terms related to genre, medium, and aesthetic judgment changed between 1905 and 1925 in a sample of book reviews published in the US periodicals. In the exploratory phase of this project, terms seemingly related to categorization are identified and divided into “feature families.” In the confirmatory phase, feature family terms are analyzed for their relatedness to one another and then evaluated for how well they predict book review dates. This comparison is conducted using a featurization method called Word Mover’s Similarity Centroid Regression, which adapts the idea of Word Mover’s Distance for a regression task. The medium feature family proved to be the most predictive of a review’s publication date, followed by judgment terms, and then genre terms.