Assessing the Sentiment of Book Characteristics Using Machine Learning NLP Models

Pawel Drozda, Krzysztof Sopyla

Abstract: This paper presents a new approach for processing the entire books with Natural Language Processing algorithms. In particular, we proposed methods to evaluate books in terms of assessing the intensity of the book’s soft features, such as fantastic, touching, suspenseful, etc. Using Bag of Words and TF/IDF, we embedded books and conducted classification experiments to determine the most appropriate parameters for classifying the intensity of features. The obtained results showed, that in the considered problem the Random Forests algorithm fitted the best, achieving accuracy of 95% and F1 measure of 89%. The evaluation also included the selection of the best converter and data aggregation method.