Research On Recommendation Algorithm In Event-based Social Network

With the rapid development and popularization of the Internet,cyberspace and the real world are merging continuously.Simply online interaction through the social network can no longer meet the needs of users.To combine virtual online social relationships with offline actual face-to-face communication,the Event-based Social Network has born.EBSN not only provides users with a platform for online communication,but also helps users coordinate and organize offline social events.As well as the traditional social network platform,EBSN is facing the severe problem of information-overload.It is difficult for users to find out the events which they are most interested in from massive data.Therefore,to improve the efficiency of EBSN and enhance users' experience and loyalty,it is of great significance to study recommendation algorithm of the EBSN.Because each event is brand-new and lacks historical information from users,recommendation in EBSN network faces more serious problem of "cold start" than traditional social network.The direction of solving this problem is to excavate deeper social network features of users and events,to improve the performance of recommendation.Therefore,this thesis proposes a recommendation algorithm based on deep neural network of social event.The main research work of this thesis includes:(1)Building up the dataset.Through web crawler and the API provided by Meetup,this thesis collects the data of social events,groups,users,etc.from Meetup.After building up the dataset,this thesis analyzes the network structure of the dataset,and proves that our dataset can provide the data-foudation for furture research on EBSN.(2)Studying the feature representation method of four factors affecting the effect of the recommendation algorithm.The user nodes in the network are presented as vectors through graph embedding algorithm,the vectors are the social factor.The latitude and longitude of the events are divided into regions by the DBSCAN algorithm based on the spherical distance,and the geographical factors are generated through One-hot encoding based on the regions.According to the users' habits,this thesis divides the time to fragments,and the temporal factors are generated through One-hot encoding based on the fragments.Finally,this thesis uses latent semantic analysis to generate semantic vectors for the event and user preferences of semantic.(3)Proposing the recommendation algorithm based on deep neural network for social events in EBSN.First of all,this thesis models the targeted users and the events to be recommended from social,geographic,temporal and semantic aspects.And then,this thesis obtains user's and social event's feature representation vectors,which are the input of the deep neural network.The algorithm obtains more advanced features of user and social event through the powerful representation ability of deep neural network.The scores between the targeted user and the events to be recommended are calculated by the multiply of the advanced features and the several events with the highest score is recommended to the targeted user.(4)Verifying the algorithm proposed in this thesis.Firstly,the output feature dimension of deep neural network is discussed,and the best output feature dimension is selected by considering both the result of recommendation and the performance of the algorithm.Secondly,the effects of four influencing factors of social,geographic location,time and semantics on the proposed algorithm are studied.Finally,the social activity recommendation algorithm proposed in this thesis is compared with other recommendation algorithms.The results show that the performance of the recommendation algorithm proposed in this thesis is better on both datasets.