Downstream scientific informativeness of generated dataset sets

Determine whether dataset sets generated by FINALLY produce different or more informative experimental conclusions than established dataset-selection practices, including frequently used benchmark selections and selections observed in published recommender-systems studies.

Background

The thesis establishes technical constraint compliance and alignment with the implemented diversity objectives, but it does not evaluate whether those properties improve recommender-systems research outcomes. The authors propose comparing equally sized diverse, non-diverse, and Random selections with conventional benchmark selections and applying the same algorithms and evaluation protocol to all sets. Whether performance-oriented dataset-set selection reveals additional empirical insight remains unresolved.