Mining Goodreads. A Digital Humanities Project for the Study of Reading Absorption

We present our method and interim results of the “Mining Goodreads” project, aimed at developing a computational approach to measure reading absorption in user-generated book reviews in English. A team of eight people (three supervisors and five annotators) have joined skills from the fields of empirical literary studies, natural language processing, and digital humanities, with the goal of producing a gold-standard annotated dataset and strengthening the theoretical framework of reading absorption. Annotation of more than 800 texts showed the difficulties in finding an agreement in the tagging of sentences. However, through more than one year of work in strict collaboration, the team reached some substantial improvements: inter-annotator agreement increased through seven annotation rounds, while machine learning approaches were applied on the annotated corpus, producing promising results.