Evaluating Recommender Systems Effect on Content Diversity: An agent-based framework
Document Type
Article
Publication Date
Fall 2020
Abstract
Digital markets depend on recommender systems that facilitate interactions among suppliers, distributors, and consumers, ultimately increasing sales volume and allegedly increasing user utility. Beyond this operational cornerstone, recommender systems also have a passive role in how these markets organize and behave (e.g., funneling consumers into few suppliers or promoting obscure products and services that can better satisfy consumer needs). Te potential effect of recommendations appears to be larger on cultural or entertainment and media industries, where a product’s uncertainty is usually high. As such, cultural diversity and market concentration on content platforms (e.g., YouTube, Spotify) are susceptible to the effect of recommendation system algorithms. Te study of diversity has been a focal topic for individual recommendation optimizations, but little attention has been given to aggregated measures of diversity. Previous work on this area states that collaborative-based recommender systems have an impact on sales diversity. I expand on this, presenting an agent-based model to test the impacts of recommender systems on cultural markets. Te model offers a framework to estimate the effects of state-of-the-art content-based and collaborative filtering algorithms on diversity. Early results confirm previous work. Next steps include the use of machine learning algorithms, social influence, and adaptive behavior of users. Current and future work will provide useful insights for marketing modeling, content platform policy, and the use of recommenders in market design.
Recommended Citation
Lhuillier, E.
(2020). Evaluating Recommender Systems Effect on Content Diversity: An agent-based framework. Journal on Policy and Complex Systems, 6 (2), 37-52.
Publication Title
Journal on Policy and Complex Systems
Start Page No.
37
End Page No.
52
ISSN
2372-8590
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License.
DOI
doi: 10.18278/jpcs.6.2.4