Judge a Book by its Cover: A Multimodal Approach to Book Genre Prediction

In today's visually driven market, book cover design plays a crucial role in conveying a work's narrative and thematic essence. However, conventional recommendation systems have largely overlooked the valuable information embedded in cover designs, particularly regarding genre classification. Recognizing a book's genre from its cover is challenging due to the subtle and complex interplay of design elements. In this paper, we propose a multi-modal approach that integrates both visual and textual features extracted from book covers to predict their genres. Our method employs two complementary image embedding models-designed to capture both local and global visual information-alongside a text embedding module that incorporates cover text. We evaluated the proposed approach on a dataset comprising 57,000 books spanning 30 genres. Experimental results demonstrate that our method achieves a Top-1 accuracy of 52.98 % and a Top-3 accuracy of 75.27 %, outperforming state-of-the-art techniques and underscoring the potential of multi-modal integration for genre recognition.