Abstract
Methods for extracting users' interests from their behaviors in the real world and on the Web have been studied. Radio Frequency Identification (RFID) is a sensor which has been employed in logistics for tracing the movements of products, and can be extended to a tool for obtaining the behavioral data of human(s). If RFID is used in a bookstore, pick-up behaviors and browsing time are to be obtained. In this paper, we hypothesize that the longer a book is picked and read, the stronger the user is interested in the book. We propose a system for extracting keywords representing user's interests from his/her behavioral data, i.e., the history of browsing books stored in bookshelves. The proposed system identifies books that have been browsed by a user, and extracts keywords that appear frequently in the books. Then the system weights each keyword using the browsing time of books including the keyword. The system finally outputs keywords of high weights as the indices of the user's interest. The experimental results support our hypothesis, i.e. the keywords obtained by the proposed system represent users' interests more precisely than previous indexing methods.