Diversity in Recommender Systems

Recommender systems play a vital role in our digital lives, supporting users in the exploration of vast content landscapes. Diversity has emerged as a cornerstone in broadening users’ horizons while simultaneously boosting companies’ bottom lines. It refers to the variety and heterogeneity of recommendation lists provided to users, aiming to offer a broader range of options and solve problems inherent to solely accuracy-focused recommender systems. This thesis offers a comprehensive theoretical examination of diversity’s desirability for different stakeholders, its varying definitions, intricacies of its measurement, and methods to augment recommendation diversity. Through empirical experiments, four widely-used recommender systems are evaluated in a top-k music recommendation scenario to give insight into their performance concerning diversity. Results revealed EASE as the best-performing model, especially regarding the trade-off between accuracy and diversity.