Event Recommendation System

Business Objective

The objective is to maximize meaningful event attendance while growing the events ecosystem and maintaining platform health. "Meaningful" means users actually attend events they RSVP to and have positive experiences, not just clicking "interested" and forgetting. We're optimizing for real-world engagement, not just digital interactions.

There's a critical difference from content recommendation: events have stakes. If we recommend a video and the user doesn't like it, they scroll past. If we recommend an event, they might travel across the city, pay admission, clear their calendar, and have a bad experience. The precision requirements are higher, and the cost of false positives is greater.

Growing the events ecosystem means helping event creators reach their audiences. A concert with 50 empty seats or a community meetup with no attendees is a failure for both creators and platform. We need to match supply (events) with demand (interested users) efficiently.

The time-bound nature creates urgency. An event next weekend needs recommendations this week, not next month. Ranking needs to account for temporal relevance, events happening soon should surface for users who might attend, but not so early they forget or so late they can't plan.

Location dependency is critical. Recommending a local community event to someone 500 miles away is useless. But for major concerts or conferences, people might travel. The model needs to understand which events are worth traveling for and which are hyper-local.

ML Objective

From an ML perspective, this is a ranking problem with unique constraints. Given a user and thousands of upcoming events, we need to rank events by the probability the user will: express interest ("interested" or "going" RSVP), actually attend (show up to the event), and have a positive experience (engage, return for future events).

The multi-stage nature of engagement creates a funnel: interest → RSVP → attendance → satisfaction. We need to predict all stages because optimizing only for interest clicks would lead to many low-commitment RSVPs without attendance.

The temporal aspect is key. We're not predicting "would this user ever be interested in this type of event?" but rather "would they attend THIS event on THIS date?" Time-to-event matters enormously. The optimal time to show an event is typically 1-2 weeks before for most event types, giving users time to plan but maintaining urgency.

Location is a hard constraint for most events. We need to filter candidates by feasible distance and model willingness to travel, which varies by event type, user patterns, and event scale.