The aim of this project is to explore the effect of the emotions expressed in fictional stories on the emotions experienced by their readers. More than 400 books from 9 different genres and their corresponding reviews from Goodreads are used. Sentiment analysis is used to calculate three different types of sentiment values: the average book sentiment, the average review sentiment, and the emotion story arc of each book. Three different methods are used, namely, a dictionary-based approach, a transformer-based approach, and a vector-space model approach called SentiArt. The plot type of every book is defined by clustering the emotion story arc using k-means with Dynamic time warping Barycenter Averaging (DBA) as the distance metric. Hypotheses are tested using linear regression models (ANCOVA) to analyze the covariance between the sentiment values of books and reviews.

On 29 June 2022, Srishti presented her project at the “Library Lunches with a taste of DH”, a series of talks organised by the Faculty of Arts and Philosophy at Ghent University. The full video is available below while the slides can be found here.