Research And Implementation Of Graph Attention Networks Based Group Event Recommendation Algorithms

Event-based Social Networks(EBSN)have drawn more attention in recent years,which link the online and offline social relationship of users.However,the enormous events in EBSN make it difficult for users to find target events.Event recommendation systems appear as an effective solution to alleviate information overload.Due to the inherent characteristics of events,event recommendation usually faces cold-start problem.Hence,it's necessary to fully exploit the abundant context features of events.Since human beings are gregarious and group activities are indispensable in people's daily life(e.g.,traveling,dining out and going to the cinema),group recommendation has drawn more attention in recent years.The theis studies graph attention network based group event recommendation as follows:(1)A new neural collaborative filtering model NCFE for event recommendation is proposed.NCFE extracts the abudant context features of events in a fine-grained way and incorporates attention mechanism to assign weights for them.A feature interaction layer is designed to capture the relationship between users and events.The experimental results on two cities in Meetup dataset show that NCFE achieves better recommendation performance.(2)A User-Difference Attention(UDA)model that explicitly simulates the comparisons between group members by relational attention is proposed,which is different from previous single user based models.Each user is compared with all other users under the guidance of the target item,then a multi-layer perceptron is exploited to add nonlinear transformations.Several User Relational Kernels(URKs)are proposed to simulate different types of relation during group decision making.Extensive experiments have been conducted on three public datasets.The results show that UDA significantly exceeds the state-of-the-art competing methods.(3)A graph attention network based group event recommendation model GAGE is proposed.GAGE uses graph graph convolutional network and attention mechanism in EBSN to capture the topology structure and learn the representation of nodes.A feature interaction layer is designed to learn the preference of groups on different context features of events.The experimental results on two cities in Meetup dataset show that GAGE achieves the state-of-the-art performance.(4)A concert recommendation system is designed and implemented,where NCFE and GAGE are deployed.The system is able to recommend concerts for both individuals and groups,which improves user experiences and commercial profits.