1. Introduction
The effective use of library loan records for generating recommendations has been actively discussed among librarians and library and information science researchers. One method is to recommend books to users based on loan records. Some studies have proposed approaches for implementing this method. However, most of these methods use only the information from loan records. Book titles and the Nippon Decimal Classification (NDC) categories that have been assigned to the books have not been used. We contend that book titles, NDC categories,
and the outlines of books from the BOOK Database are important additional clues that can be used for the purpose of formulating effective book recommendations and that the optimum combination and weighting of these additional clues can be determined through machine learning.