Multi-Feature Based Social Events Recommendation

In recent years, event-based social networks prospered. Event-based social networks mainly aim at helping users organize offline activities. On this kind of social network, users can create and join groups, and members of groups can organize offline activities. As the data size on event-based social networks increasing, recommendation algorithms are needed to help user get over the selection problem. Events have their own characteristics, for example, that events are allowed to participate in only one, and have no historical rating records, and that there are both online and offline social networks. Due to these characteristics, traditional recommendation algorithms cannot be applied directly, thus we need to innovate the recommendation algorithms.In this paper we introduce a event recommendation algorithm we’ve designed and implemented, whose job is to generate events’features, and calculate the matching score based on these features. The proposed algorithm composes of two parts: feature extraction and learning to rank. In feature extraction, six features are considered, including the user preference, collaborative filtering based preference, area popularity, social influence of events and geographical affinity. We applied the ideas of collaborative filtering recommendation and content-based recommendation in feature extraction. In learning to rank, we use pair-wise learning to rank to calculate the weight of each feature, thus integrating all features into a linear combination.In order to verify the proposed algorithm, we’ve collected the dataset and run many experiments. Our experiment results show that the proposed recommendation algorithm performs well, and outperforms the algorithms compared. Besides, for confirming the usability of the proposed algorithm, we’ve implemented an application for event recommendation. Under the real scenario, this application runs well.