5-01 - Practical Evaluation of Repeated Recommendations in Personalized Music Discovery
Abstract: Studies have shown that repeated exposures to novel songs cause an increase in a person’s memory and liking. These studies are commonly verified through self-reporting emotion-based surveys. This paper proposes the “retention rate” as an additional parameter for evaluation. The “retention rate” is one at which the listener revisits the novel items. The authors hypothesize that when a person listens to novel (i.e., both unfamiliar and interesting) pieces of music, the retention rate will be proportional to the number of times the discovery engine suggests the pieces to her, as long as they remain novel. The authors have tested the hypothesis through a six-week human-subject experiment that simulates a real-world listening environment and a follow-up survey. During the experiment period, each subject received, through Discover Weekly in Spotify, suggestions for novel songs up to three times and provided evaluation. One month after the evaluation experiment, the human-subjects answered whether they had revisited the novel songs. Through the analysis of the response and survey data, the researchers conclude that the more times a listener is exposed to a song during the discovery process, the more likely she is to return to the song.