Abstract
Recently, radio listening styles have changed, and there are listeners who listen favorite programs carefully instead of listening while working.Needs for radio program recommendation are increasing, but there are few studies of radio program recommendation, and suitable recommendation methods are unknown.This study collects users' interests, followers/followees, and listened radio programs from Twitter, and validates accuracy of recommender systems applied collaborative filtering and Random Walk with Restart (RWR).An account, who tweeted with hashtags specified in a radio program website, are treated as an observed listener of the program.This study collects one year's likes to tweets and followers/followees on 1000 Twitter accounts who listened 16 programs of All Night Nippon series, Nippon Broadcasting System in December 2021.According to the radio program recommendation experiment, it is confirmed that the interest based RWR and collaborative filtering achieve the best recommendation accuracy for users without listening histories and for users with histories, respectively.
