2026/04/22 by Marijn Koolen, Joris van Zundert, P. Boot +2 · 1 voice
Arts and Humanities · Computer Science · #Authorship Attribution and Profiling #Media Influence and Health #Sentiment Analysis and Opinion Mining
paper · doi:10.26083/tuda-7995
openalex publication_date 2026/04/22 · openalex created_date 2026/05/06 · openalex updated_date 2026/07/31
With the availability of massive amounts of data, online reviews have become a promising source for detecting readers’ engagement with fiction. Yet identifying expressions of response to reading (such as emotions, evaluations, and feelings) remains challenging for automated analysis, given the unstructured nature of reviews. This study proposes a method for categorizing review content in order to distinguish between sentences referring to readers’ experiences and those addressing other book-related aspects, such as plot, characters, or author. We first designed an annotation schema, after assessing previous classifications. Then, we manually annotated according to our schema the sentences from 1,400 Dutch fiction reviews. Subsequently, we trained a robBERT2023 classifier to automate review sentence classification and applied it to a corpus of 670,751 reviews. Our analysis allows us to draw conclusions on the composition of online reviews and correlations among their components.