There it suggests events that its users might be interested in based on those that they’ve attended in the past, as well as information from their social graph. Sharma said the company is using a big Hadoop cluster to process the data, going through about 1.5 terabytes of data each day.īut it seems to be working: Right now, Eventbrite’s big recommendations push comes in the form of email newsletters, which are still a very powerful tool for e-commerce. In both cases, Eventbrite is crunching a huge amount of data to power these recommendations. To do so, it puts events into up to 42 different categories, and from there can determine what types of events a user might want to go to, based on common groupings of events.
Extend that out another degree, to friends of friends, and it’s looking at more than 1.25 trillion.īut Eventbrite is also drawing recommendations from implicit information - including the types of events that they’ve already attended. When you consider that Eventbrite is doing this for 20 million users, that means it’s processing 5 billion pieces of information. On average, each attendee has an average of 250 friends. The company’s recommendations are drawn from two main sources: On one hand, Eventbrite can make explicit recommendations based on data that users have shared on Facebook, including Likes and Interests, as well as their social connections. For each of those users, it can provide an average of 60 events to recommend, depending on how much data Eventbrite has on their interests and other events they’ve attended. Today, Eventbrite is providing recommendations for about 20 million of its users, Director of Data Engineering Vipul Sharma told me. In December, it launched a recommendation system that would offer up events its users might want to attend, and now, about seven months later, it’s sharing some data about how that system is performing. Online ticket sales startup Eventbrite is always looking for new ways that it can boost attendance at various events that are run on its platform.
0 kommentar(er)
