Explanations in Music Recommender Systems in a Mobile Setting
MetadataShow full item record
- Master theses 
Every day, millions of users utilize their mobile phones to access music streaming services such as Spotify. However, these `black boxes’ seldom provide adequate explanations for their music recommendations. A systematic literature review revealed that there is a strong relationship between moods and music, and that explanations and interface design choices can effect how people perceive recommendations just as much as algorithm accuracy. However, little seems to be known about how to apply user-centric design approaches, which exploit affective information to present explanations, to mobile devices. In order to bridge these gaps, the work of Andjelkovic, Parra, & O’Donovan (2019) was extended upon and applied as non-interactive designs in a mobile setting. Three separate Amazon Mechanical Turk studies asked participants to compare the same three interface designs: baseline, textual, and visual (n=178). Each survey displayed a different playlist with either low, medium, or high music popularity. Results indicate that music familiarity may or may not influence the need for explanations, but explanations are important to users. Both explanatory designs fared equally better than the baseline, and the use of affective information may help systems become more efficient, transparent, trustworthy, and satisfactory. Overall, there does not seem to be a `one design fits all’ solution for explanations in a mobile setting.
Revised version: some spelling errors corrected.