Group-Aware Fair and Diverse Recommendation via Dual-View Preference Modeling and Rank-Sensitive Re-Ranking: Evidence from Real Travel Groups

Abstract

Group recommender systems must reconcile collective relevance with divergent member preferences. This study introduces Group-Aware Fair and Diverse Decisions (GA-FDD), a transparent decision-layer framework integrating member-derived and direct-group-history relevance signals with rank-sensitive member utility, model-predicted within-group fairness, and list diversity. The principal methodological contribution of GA-FDD is the explicit separation of these two relevance views, while fairness is determined exclusively from member-level predicted satisfaction rather than collective group history. GA-FDD is evaluated on an established real-world travel-group dataset comprising observed memberships and individual and collective interaction histories using a group-disjoint validation and final-test protocol complemented by sensitivity, robustness, and group-level cold-start analyses. The proposed method is benchmarked against representative approaches spanning conventional aggregation, fairness- and diversity-aware re-ranking, direct-group collaborative models, and recent neural group recommendation. Relative to ConsRec, a recent neural multi-view consensus model evaluated under the same protocol, GA-FDD preserves a comparable hit rate while yielding substantially higher model-predicted minimum member satisfaction and Jain fairness, alongside comparable list diversity. While this comparison provides a relevant neural baseline, it is intended as a targeted comparison rather than an exhaustive benchmark of recent neural methods. Overall, the findings support GA-FDD as a strong cross-objective compromise and configurable decision layer for settings in which relevance, member-level fairness, and diversity must be jointly governed rather than optimized in isolation.