Logistic Regression Modeling for Predicting Commercial Publication of Online Novels Using Only Attribute Information

The growing influence of User-Generated Content (UGC) in the digital content industry has increased the need for reliable methods to assess its commercial potential. This study focuses on predicting the likelihood of commercial publication for novels posted on “Shosetsuka ni Narou,” one of Japan’s largest online writing platforms. We constructed a binary classification model using logistic regression based solely on attribute data of the works, excluding textual elements such as titles, synopses, and main content. Work data were collected via API and web scraping, and filtered by genre, character count, and reader evaluation to ensure relevance and quality. The analysis identified several key features influencing publication outcomes. In particular, works in the “Isekai” genre showed a higher likelihood of being published, while those with BL/GL elements or reincarnation/transportation tags were associated with lower publication rates. To convert predicted probabilities from the logistic regression model into binary classification outcomes, we adopted a decision threshold of 0.2. This setting prioritized high recall, yielding a Recall of 0.79 on test data and enabling the identification of many potentially publishable works, albeit with lower Precision. These findings demonstrate the feasibility of early-stage publication prediction using non-textual metadata and highlight how platform-specific preferences and tagging behaviors affect publication trends. The proposed framework provides practical support for editorial decision-making and is adaptable to other UGC platforms with appropriate modifications.