Abstract
Recently, there are many reading support web services, which support users' communication about their impression of books. To facilitate smooth communication between users, the function supporting to meet other users with similar interests is necessary to these services. In this paper, we propose a method for discovering users who have similar interests, "NDC Tree Profiling", for reading support web services. NDC tree profiling generates tree forms of user profile using numbers of Nippon Decimal Classification (NDC) of the books they read. Then, comparing each profile, our method could discover users who have similar interest for reading. The result of the evaluation experiment, our method was statistically superior to random method. Also our method is more accurate than the method based on the number of common books, and the method based on vector space model using TF-IDF, but this difference was not statistically significant. Although sufficient accuracy improvement was not completed, the NDC tree profile proposed by this research has a layered structure, and has a possibility that a user's concern can be more delicately caught by adjusting the dignity for every class. More effective similar reader discovery may be realizable by performing further adjustment.
