ScholarWorks@숙명여자대학교

초록

Bestsellers are the most common way for readers to choose books, and for this reason, the prediction and selection of bestsellers is an important marketing strategy indicator in the publishing market. In this study, we propose a model that predicts whether or not to remain in the top 200 bestseller rankings and sales index sections using metadata from bestsellers, and compare and evaluate the performance of various machine learning algorithms. To this end, monthly bestseller data on the Yes24 site were crawled and collected, and appropriate preprocessing was performed for each data attribute. Various classification algorithms were used to predict whether to maintain the ranking, and as a result of finally evaluating the prediction performance of each algorithm, the accuracy of MLP, CatBoost, and random forest was high. This study is meaningful in that it comprehensively compared the performance performance of various classification algorithms for predicting whether to maintain the bestseller ranking. However, in models that rely on the number of reviews and ratings as limitations, it was difficult to overcome the cold start problem in new books that lacked data, and the need for follow-up supplementary research is proposed.

키워드