This project aims to predict the future success of newly-released novels within the SFF (science-fiction and fantasy) genres. In particular, we focus on predicting if a given new release will be nominated for a prestigious SFF book award: the annual Hugo Award for Best Novel and Locus Awards for Best Fantasy and Best SF Novel. We combine data from ISFDB, an online SFF novel database, supplemented by user-generated content tags and novel synopses from the Hardcover reading social media app API.

We explore the impact of features such as authors’ past award success and publisher prestige, as well as novel similarity and distinctiveness through their user-generated tags; we also encode book synopses using natural language processing methods as an additional feature.

We investigate classifier and LearnToRank models to predict nomination outcome, and find that LearnToRank models have predictive power above our baseline models, even in a highly imbalanced dataset.