Use of Library Loan Records for Book Recommendation

To determine the most effective book recommendation method for libraries, we conducted a recommendation experiment using (1) collaborative filtering based on the library loan records, (2) association rule mining based on the same data, and (3) Amazon. The library loan records of a certain university library for the period 2006 to 2011 were used. We recommended books to 33 students and asked them to describe the books' level of interest. The results show that books recommended by collaborative filtering were least favorably evaluated, followed by those recommended by association rule and Amazon. Collaborative filtering carries the risk of breaching users' privacy and its computational costs are higher compared to those of association rule mining. Therefore, if we recommend books based on library loan records, association rule mining should be adopted instead of collaborative filtering. In addition, given the fact that the recommendations by Amazon were most favorably evaluated, utilization of Amazon should also be considered.