Towards Fairness-aware Multi-Objective Recommendation Systems

Recommender systems (RSs) have been extensively developed to provide personalized recommendations for the end users from the near-infinite options on the internet. Accuracy is the main focus of any recommender system. Nevertheless, the already existed accuracy-focused RSs have largely ignored to consider different users' rating behaviour. The utilization of users' rating behaviour can significantly help to improve recommendation accuracy and fairness. However, only focusing on the accuracy of recommendation can lead to the development of RSs reinforcing only popular items and ignoring other important objectives (e.g. long-tail inclusion or diversity) that have a significant impact on the overall quality of a recommendation system. Consequently, the focus of the scientific community on RSs have been recently shifted to cover a wider range of objectives. The incorporation of multiple objectives in the recommendation process is referred to as a Multi-Objective Recommender System (MORS). This multi-objective design pattern poses a key challenge for recommendation fairness towards users, providers, and items. The other challenge that should be considered in recommendation is that users' long-term interactions are often not as equally important as their recent preferences since users' interests change over time. This is known as the study of the Session-based Recommendation System (SRS). Along these lines, the goal of this thesis is to address the above-mentioned issues and advance the scientific understanding of fairness-aware recommendation systems with various objectives. The main objectives of this thesis are (i) users' rating credibility calculation in an accuracy-focused RS, (ii) long-tail inclusion in a fair MORS, (ii) personalized diversity in a fair MORS, and (iii) personalized diversity in a fair multi-objective SRS.