Automatic Classification of Reference Service Records

The National Diet Library in Japan maintains a database of reference service questions and the answers given to them (henceforth, reference service records). The questions are submitted to public and university libraries in Japan by users, and the answers are given by the libraries. To improve the findability of these records, we propose a method for automatically assigning Nippon Decimal Classification (NDC) codes to them. Although some studies have been conducted on classification of the reference service records, their precision is unsatisfactory. One reason for this is that these methods depend only on the texts of the reference service records. Many reference service records contain book titles that the reference librarians believe will be helpful to the questioners. The NDC codes of these books might be useful for classification. Based on this background, we propose to automatically (1) extract the book titles in the reference service records, (2) submit them to the NDL-OPAC, (3) obtain the NDC codes of the books, and (4) input them into a support vector machine (SVM) as features together with the texts of the reference service records, and then classify them. A total of 62,884 records were used. For the first digits of the NDC codes, our method achieved a 53.4% precision. For the first and second digits of the NDC codes (i.e., classes and divisions), our method achieved a 46.1% precision. These figures are significantly higher than those of the preceding studies.