Enhancing book recommendation with automated genre mining

Abstract

Genre classification of books plays a pivotal role in enhancing the overall user experience in the rapidly evolving digital landscape of literature consumption. As the number of e-books and digital titles continues to grow exponentially, readers often face an overwhelming amount of content, making it increasingly difficult to discover books that match their interests. Accurate genre classification not only simplifies this discovery process but also serves as a foundational element for personalized recommendation systems, intelligent content organization, and improved search relevance. These systems rely on genre tags to guide readers through curated suggestions, thereby increasing engagement, satisfaction, and the likelihood of continued platform usage. Furthermore, for publishers and platform designers, effective genre classification provides insights into emerging trends, audience preferences, and market dynamics, supporting decisions related to marketing, acquisition, and catalogue management. This thesis investigates the task of automated book genre classification through two complementary yet independent modalities: visual semantics derived from book cover images and textual semantics mined from book blurbs and user-generated reviews. Each modality addresses specific challenges in genre inference and serves distinct practical use cases. The visual approach is particularly beneficial when textual metadata is unavailable, unreliable, or intentionally misleading, while the textual analysis is better suited to capture deeper thematic and narrative content.