Proposal of a New Recommendation System that Addresses “Personalizability”

Abstract

Users’ accessibility to a tremendous amount of information has increased because of the recent development of various media platforms and the resulting spread of the Internet. Alternatively, however, users are finding it increasingly difficult to locate meaningful data. “Recommending” information has become a popular field of study. A variety of recommendation systems are currently in use. “Accuracy” is the main performance index used to assess these systems. However, other evaluation indexes have been proposed. One particular index, “Serendipity,” provides a description of each recommended item’s newness and states whether each item contains an element of surprise. In this paper, we provide a definition of “Personalizability,” one of several factors that must be considered during an evaluation of “Serendipity.” In addition, we propose a recommendation method that employs “Personalizability,” and we apply the proposed method to benchmark data. We demonstrate that “Personalizability” can be used to tweak “Accuracy.” We also demonstrate how this method improves on conventional methods in “Personalizability.”