Recommendation diversity

Left alone, recommendation engines converge on the safe bet: the big-name exhibitors everyone clicks, the session titles everyone recognizes, the profiles most similar to what someone already viewed. Accuracy metrics reward this — safe suggestions get accepted — but the event suffers in ways the dashboard doesn't show. Attendees see a narrow slice of the floor and stop discovering; the long tail of smaller exhibitors, who often decide whether your show feels worth the stand fee, become invisible to the algorithm that routes footfall. Diversity is the corrective: deliberately blending expected matches with adjacent categories, newcomers, and wildcard suggestions. For organizers, this is commercial, not aesthetic — a recommender that concentrates attention on your twenty anchor exhibitors is quietly undermining renewals across the other four hundred. In practice, you manage it by asking vendors directly how their system balances relevance against exposure, whether newcomer exhibitors get a visibility boost, and by checking distribution yourself: pull the recommendations for fifty attendees and count how often the same ten exhibitors appear. The common mistake is judging matchmaking purely on acceptance rate, which structurally favours bland suggestions. One honest nuance: diversity costs some accuracy by definition — a few suggestions won't land. That trade is worth making, but pretending there's no trade is how vendors dodge the question.