Abstract
Recently, a user can access a huge amount of information with the popularization of Internet and the development of media associated with that. On the other hand, it becomes difficult for him/her to choose desirable or valuable ones. Then various recommendation systems have been studied and put into practical use to support users' selection. Though "accuracy" has been used as the evaluation of these systems so far, it is said that other evaluation factors are also needed from the aspect of the satisfaction of users. "Serendipity," which includes novelty or unexpectedness, is one of the evaluation indexes for user satisfaction. This paper defines "Personalizability" as an approach to quantify the "Serendipity," which is the degree of the specialization for each user and the index to represent how much specialized the recommended item is to the user with his/her favorite while it is unfavorite one for other users. This paper proposes a recommendation system considering "Personalizability" with the assumption that the improvement of "Personalizability" makes that of "Serendipity." This paper applies the proposed method to benchmark data. It shows that this method can adjust the balance between "Accuracy" and "Personalizability" and it is superior to the conventional method in "Personalizability."
