Investigating the efficacy of persuasive strategies on promoting fair recommendations
Date
2022-04-27
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0003-2547-124X
Type
Thesis
Degree Level
Masters
Abstract
Fairness in recommender systems has gained lots of attention, considering provider and system objectives along with end-user satisfaction. However, often there are trade-offs between the objectives of different stakeholders. Music recommender systems suffer from popularity bias, meaning that songs from famous artists are widely recommended; in contrast, new artists on the same platform struggle to attract listeners. However, less popular providers might not satisfy users as much as widely-known providers; therefore, user satisfaction might decrease significantly. Consequently, there is a need to explore methods to promote recommendations from less-known providers. Previous studies have shown that explanations and persuasive explanations are beneficial for increasing user acceptance of recommended items. However, there has been little work investigating explanations for a fairness objective. This research is focused on the effect of persuasive strategies for promoting items included for the recommender's fairness objective in a music platform, highlighting which persuasive strategies can be used to create influential persuasive explanations. Results show empirical evidence of higher user satisfaction for the items accompanied by explanations. The findings of this thesis could guide the user interface design of multi-stakeholder recommender systems leading to better user satisfaction. Moreover, the impact of different demographic features and personalities on the ratings of songs from new artists is explored. Based on our results, users with different demographic characteristics and personalities are receptive to distinctive persuasive messages. This information provides a better understanding of the participants' behaviour, leading to personalized guidelines for designing persuasive fair music recommender systems. Furthermore, users' perception of persuasive strategies that they are susceptible to is compared with the actual persuasive strategies that the users were influenced by based on the rating users provided to the songs from new artists and persuasive messages individually. The comparison of the ratings yielded that users correctly identified influential and uninfluential persuasive messages with 38.25% accuracy. Scarcity was the most underestimated method; the users' perceived persuasiveness of this method was very low. However, the ratings of songs from new artists showed that this method affected users' ratings. This result shows that personalizing persuasive strategies solely based on the users' opinions about their receptiveness to the persuasive strategies might not reflect the true power of persuasion, at least in music recommendation.
Description
Keywords
persuasive technologies, fair recommendation, explainable recommendation
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science