Study of event recommendation in event-based social networks

Abstract

[en] Recommendations are in our every day life: streaming services, social media, web pages... are adopting and using recommender algorithms. Recommendation algorithms benefit both parts: clients can find more easily products that they like, and the companies make more benefits because clients use their services more. The recommendation problem presented in this work is a non-traditional variant of this problem as it recommends events. Events, unlike books or videos, cannot be recommended in the same way, because users cannot rate an event until the day it happens, and then no new users can rate it again after that. This magnifies a problem called “cold start problem” where every new event has no ratings, which greatly complicates the recommendation problem. This work studies Event Recommendation for a social media called Meetup 1 where users can attend a selection of events created by the community. Although users do not leave a rating of the event, we have a signal called RSVP 2 , which is a non-obligatory mark on whether the user has the intention to attend the event or not. In this work we will be exploring how different recommender algorithms perform to recommend events based on RSVPs and also propose three new algorithms. The analysis will be done with 5 datasets extracted from Meetup during the months between November 2017 and April 2018. The results show that hybrid versions containing collaborative and contextual-aware algorithms rank the best among all the algorithms tested.