The Challenge of Heterogeneity in Recommender Systems

Recommender systems (RS) have become integral to everyday digital life, helping users navigate vast information spaces. While early RS research focused on accuracy, the field has since broadened to tackle multiple "axes of heterogene- ity"-challenges arising from optimizing conflicting objectives, handling multimodal signals, accounting for diverse user preferences, or balancing stakeholder needs. This thesis focuses on two of these axes: (1) optimizing multiple, often conflicting quality criteria (e.g. accuracy, diversity, or novelty of recommendations), and (2) modeling preference heterogeneity (whether it is different propensities of indi- viduals towards recommendation quality criteria, or different preferences of the members of the same group). These themes are explored through two domains: multi-objective RS (MORS) and group RS (GRS). In MORS, the thesis presents novel algorithms, evaluation tools, and studies the aspects of user control and multi-criteria trade-offs and their perceptions. In GRS, it contributes towards al- gorithmic improvements and explores methodological aspects of offline evaluation, highlighting common issues and proposing mitigations. The thesis also reflects on conceptual links between MORS and GRS. Together, these contributions aim to make RS more flexible, transparent, and...