抄録

◎Christina Natalia Pakpahan・Wataru Kameyama（早大）

Most of audiovisual-program recommendation systems focus on ratings that explicitly given by users. However, these ratings somehow are hard to collect because users usually feel lazy to give their ratings. By utilizing user-logging data, we can consider what users' interests are based on watching duration. Besides, users' watching activities, i.e. a factor of users' similarity, are to be considered to classify them into some groups for better recommendation. The experimental results shows slightly better improvements in terms of prediction accuracy as well as diversity by applying these features to the ever-proposed "Watch-Flow Algorithm".