Research On Event Recommendation For Event Based Social Network

In recent years,Event-based Social Networks have become more and more popular,such as Meetup,Facebook Events and Douban Events.These social networks allow individuals or organizations to create,publish and manage various types of events.Users can not only communicate and share experiences online,but also make new friends and expand social relationships by participating in offline events.Effective participation in events not only enriches the user experience,but also enables the organizers of events to receive more attention.Accurate event recommendation plays a guiding role in organizing or participating in events.Therefore,the research on event recommendation has practical significance.However,as event based social networks are heterogeneous and complex networks,and events are different from general items,traditional recommendation algorithms are difficult to directly apply to the field of event recommendation.According to the characteristics of event recommendation,this paper proposes an event recommendation method based on feature extraction and fusion technology.The work of this paper mainly includes the following three aspects:(1)A social event recommendation model was designed.The social event recommendation model was designed based on the overall consideration of various factors that may affect the recommendation results.(2)A recommendation method for social events is given.The method firstly models and calculates the similarity feature values of user and event from the time factor,spatial factor,content factor and social factor that influence the participation behavior of users.Then,candidate event sets are respectively generated according to different similarity feature values,and all candidate sets are merged and superimposed to obtain a total candidate set.Finally,a deep ranking network is designed based on the idea of pairwise learning to rank.In order to improve the recommendation effect,the ranking evaluation index NDCG is incorporated into the loss function;in order to accelerate the model training process,the mini-batch gradient reduction algorithm is improved.The ranking network obtained by the training can calculate the comprehensive score of the event according to the feature values extracted in the feature extraction stage,and the recommended list can be obtained by using the trained ranking network to sort the events in the candidate set.(3)The effectiveness of the proposed event recommendation method is verified.A large number of experiments were carried out on the Douban Events dataset.The experimental results show that compared with the single feature recommendation method and the selected baseline methods,the feature fusion method can obtain better recommendation effect;the feature fusion method has better recommendation effect for users and organizers with higher activity.