TY - JOUR
T1 - ‘This book makes me happy and sad and I love it’
T2 - A Rule-based Model for Extracting Reading Impact from English Book Reviews
AU - Koolen, Marijn
AU - Neugarten, Julia
AU - Boot, P.
PY - 2022/11
Y1 - 2022/11
N2 - Being able to identify and analyse reading impact expressed in online book reviews allows us to investigate how people read books and how books affect their readers. In this paper we investigate the feasibility of creating an English translation of a rule-based reading impact model for reviews of Dutch fiction. We extend the model with additional rules and categories to measure reading impact in terms of positive and negative feeling, narrative and stylistic impact, humour, surprise, attention, and reflection. We created ground truth annotations to evaluate the model and found that the translated rules and new impact categories are effective in identifying certain types of reading impact expressed in English book reviews. However, for some types of impact the rules are inaccurate, and for most categories they are incomplete. Additional rules are needed to improve recall, which could potentially be enhanced by incorporating Machine Learning. At the same time, we conclude that some impact aspects are hard to extract with a rule-based model. When applying the model to a large set of reviews, lists of the top-scoring books in the impact categories show the model's prima-facie validity. Correlations among the categories include some that make sense and others that require further research. Overall, the evidence suggests that for investigating the impact of books, manually formulated rules are partially successful, and are probably best used in a hybrid approach.
AB - Being able to identify and analyse reading impact expressed in online book reviews allows us to investigate how people read books and how books affect their readers. In this paper we investigate the feasibility of creating an English translation of a rule-based reading impact model for reviews of Dutch fiction. We extend the model with additional rules and categories to measure reading impact in terms of positive and negative feeling, narrative and stylistic impact, humour, surprise, attention, and reflection. We created ground truth annotations to evaluate the model and found that the translated rules and new impact categories are effective in identifying certain types of reading impact expressed in English book reviews. However, for some types of impact the rules are inaccurate, and for most categories they are incomplete. Additional rules are needed to improve recall, which could potentially be enhanced by incorporating Machine Learning. At the same time, we conclude that some impact aspects are hard to extract with a rule-based model. When applying the model to a large set of reviews, lists of the top-scoring books in the impact categories show the model's prima-facie validity. Correlations among the categories include some that make sense and others that require further research. Overall, the evidence suggests that for investigating the impact of books, manually formulated rules are partially successful, and are probably best used in a hybrid approach.
KW - reading mpact
KW - Goodreads
KW - Online Book Reviews
KW - impact model
U2 - 10.48694/jcls.104
DO - 10.48694/jcls.104
M3 - Article
VL - 1
JO - Journal of Computational Literary Studies
JF - Journal of Computational Literary Studies
IS - 1
ER -