MACEDO, A. Q.; http://lattes.cnpq.br/6907208655251013; MACEDO, Augusto Queiroz de.
Resumo:
The Web has grown into one of the most important channels to communicate social events
nowadays. People plan, share and comment meetings through the Web. The event-based
social networks (EBSNs) have been created to help people meet and know each others in
a simpler and faster way. However, the sheer volume of events available often undermines
the users’ ability to choose the events that best fit their personal preferences. Recommender systems appear as a natural solution for this problem. However, differently from the classic recommendation scenarios (e.g. movies, books, restaurants recommendations),the event recommendation problem is intrinsically cold-start, there is no information about the users and items interactions in recommendation time. At first, this interaction only happens after the ocurrence of the event. And even using the
RSVPs informations (i.e. declared user intention to attend or not a future event), the recommender will still have to face its high sparsity worsened by the trend that users have to send RSVPs near the ocurrence of the events. To overcome this limitation, we propose a hybrid recommendation model that optimizes a personalized ranking of events based on the several contextual signals available in EBSNs. Besides social signals derived from RSVPs and user’s associations in online groups, we exploit the content signals from events’ description, location signals based on the users’ home and events geographic coordinates and temporal signals derived from the temporal and weekday users’ preferences related to their past events. Thorough experiments using a large crawl of Meetup.com we improved in more than 60% the evaluated personalized ranking metric with our multi-contextual learning approach when compared to a state-of-the-art event recommender from the literature.