Book Recommendation Using Machine Learning Methods Based on Library Loan Records and Bibliographic Information

We propose a method to recommend books through machine learning modules based on several features, including library loan records. We evaluated the most effective method among ones using (a) a Support Vector Machine (SVM), (b) Random Forest and (c) Adaboost, as well as the most effective combination of relevant features among (1) library loan records, (2) book titles, (3) Nippon Decimal Classification categories, (4) publication year and (5) frequencies at which books were borrowed. We performed an experiment involving 40 subjects who are students at T University. The books that our methods recommended and the loan records that we used were obtained from the T University Library. The results show that books recommended by the SVM based on features (1), (2), (3) and (5) were rated most favorably by the subjects. Our method outperforms preceding ones, such as the method proposed by Tsuji et al. (2013), and is comparable in performance to the recommendation by the website Amazon.co.jp.