Traditional group recommender systems have largely treated group members as independent sources of preferences to be aggregated. However, recent research has brought to light a paradigm shift: group interactions themselves — the ways users influence, respond to, and negotiate with each other — are becoming central modeling signals.

In this talk, we explore how explicit user interactions can be leveraged at different stages of the group recommendation pipeline to drive adaptation, consensus, and explainability. First, we present a conversational group recommender that captures social interactions in real time to infer collaboration and leadership roles, which are then incorporated into interaction-aware consensus strategies. Second, we describe a two-phase system that separates individual preference elicitation (via a conversational agent) from collaborative decision-making, supported through interactive visual explanations that surface trade-offs and preference overlaps.

These two systems embody a shift from modeling users in isolation to modeling group behavior in context. Hence, by rethinking interaction as both input and outcome, we can design group recommenders that are more transparent and engaging for users.