Group Event Recommendation Using Attention Mechanism In Event-based Social Networks

In recent years,Event-based Social Networks(EBSN)has become a popular social media service.EBSN includes not only online social interactions,but also valuable offline social activities.In the EBSN,a large number of events are released every day,which brings great difficulties to event publishers and participants.On the one hand,for event publishers,how to make the events they provide stand out from the masses of events and get user attention;on the other hand,it is also difficult for participants to find events of interest from the masses of data.Therefore,in order to help users quickly and accurately find the events they are interested in,it is very important to study event recommendation algorithms in EBSN.In EBSN,users usually participate in activities as a group,and how to recommend events of interest to the groups has gotten the attention of scholars.This paper mainly studies the group recommendation strategies based on the attention mechanism to improve the recommendation performance.Aiming at the sparseness of data in EBSN and the dynamic characteristics of group preference acquisition,this paper proposes a group event recommendation method based on knowledge graph and collaborative attention.First,it uses the knowledge graph to obtain more auxiliary information,and capture the information of the context to members of groups and events through the multilayer perceptron(MLP)and neural attention network,and obtains the embedding of the members and events under the context.Then,the embedding of users under the different context are aggregated.Finally,in order to obtain group preferences,a neural attention network is used to learn the weights of users when making decisions on candidate events,and the weights and the corresponding embedding of users are summed to obtain the group preference.The final scoring prediction is carried out through the factor machine model to realize Top-N recommendation.The performance test is carried out on the real data sets,and the results show that the proposed recommendation strategy has a good recommendation effect.Aiming at the interactive characteristics of users in groups of EBSN,this paper proposes a group event recommendation method based on self-attention mechanism.First,it learns low-dimensional embedding of users,groups,and events through the embedding layer,and learns the interaction between users through the self-attention mechanism,and obtain the similarity weight of users.Then,the similarity weight and user embedding are weighted to obtain the group's micro-preference,and the group's macro-preference is obtained through the MLP layer.the group's micro-bias and macro-preference are aggregated to obtain the group's preference.Finally,Top-N recommendation lists are achieved through metric learning.Experiments on two real data sets have proved the effectiveness of the method in this paper.