What makes a novel timeless?

Novels that are bestsellers upon publication are not necessarily popular decades later. Timeless books—popular well after publication—are a rare occurrence, standing the test of time and distinguishing themselves from the millions of books published each year. Is it possible to predict which books will be timeless, versus merely best-selling, based on word usage alone? We built two corpora of books, published between 1909 and 1923: (1) those that were best-selling (i.e., popular upon publication), and (2) those that have become timeless (i.e., popular a century later). Machine learning models were then trained to predict whether a novel was timeless or merely best-selling, based on mostly word-level psycholinguistic features. Our models were able to achieve approximately 70% classification accuracy and identified several key features of timeless books that distinguished them from those merely popular upon publication (e.g., fewer social words; more plural pronouns and words per sentence).