What can Online Book Reviews Reveal about Readers and Platforms?

Marijn Koolen, Olivia Fialho, Julia Neugarten, Joris Van Zundert, Willen van Hage, Ole Mussmann and Peter Boot

Online book reviews offer a valuable large-scale resource for studying how books affect their readers, as reviews are unobtrusive observations of readers’ opinions and descriptions of their reading experiences. At the same time, they are influenced by the platforms on which reviewers write and publish these reviews. The possibility to receive comments, likes or dislikes means people may adjust what they say to how they want to be perceived by others.

In this paper we investigate the language use of reviewers on different online book reviewing platforms, and the extent to which they reveal any impact that books have on their readers. In particular, we look at the use of personal pronouns and investigate whether reviewers relate aspects of the book and of the reading experience to themselves or to others (readers in general or readers of the review), as these may vary between different types of readers, e.g. expressive versus interpretive readers.

We compare 472,810 Dutch online book reviews from seven different online platforms with different aims (selling versus discussing books) and different social features, e.g. ratings, likes/dislikes and comments.

All reviews were parsed using the Alpino parser and syntactic information was aggregated at the level of individual reviews. We use the Reading Impact Model of Boot & Koolen (2020) to measure four types of reading impact (emotional impact, aesthetic feeling, narrative feeling and reflection) as well as the presence of seven different reading aspect categories (related to author, book, character, plot, setting, style and topic). In addition, we look at personal pronoun categories, syntactical complexity and the lexical categories of the Linguistic Inquiry and Word Count (LIWC) tool.

Our findings are that reviews across different platforms are very similar in terms of their distributions of sentence length, parse tree depth and structure, as well as on most LIWC categories and reference to book aspects.

We find that the main differences are in the length of reviews and the use of personal pronouns and that the textual characteristics of online reviews change with their length. Some platforms have many short reviews of only a few words, while others mainly have reviews of a few hundred to a few thousand words.

In longer reviews, the relative use of 1st and 2nd person pronouns is positively correlated, while they are negatively correlated with the use of 3rd person pronouns. That is, longer reviews seem to focus either on reader experiences, or on character experiences. The longer reviews on the book selling platform have more 2nd person pronouns (perhaps directly addressing the audience of potential buyers), while reviews on a platform focusing on thrillers and crime novels, which are mostly story-driven, use more third person pronouns and contain more references to the story and the characters.

These findings may indicate that these subsets of reviewer communities and reviewers relate not only to the goals of the platforms, but also to different types of readers, e.g. expressive readers who self-identify with story or characters and interpretive readers who relate to characters as others.