The Design And Implementation Of Event Recommendation System Based On Heterogeneous Social Network Relations And Topic Model

With the rapid development of social networks,a new type of heterogeneous social networks has emerged:Event Based Social Network(EBSN).Faced with a large number of events on EBSN social platform,users often face the problem of difficult selection,so event recommendation based on EBSN has attracted people's attention.Unlike other recommendation,many features of EBSN make event recommendation face many challenges:event content is a direct factor to attract user participation,but event description text is relatively short,and there are many noise words,traditional content-based recommendation can not achieve better results.In order to solve these problems,a new topic model TM-HSNT is proposed based on EBSN's heterogeneous social network features and events content text information.According to the similarity between user topics and event topics,users can recommend events of interest.The superiority of the model is proved on Meetup dataset.Finally,an event recommendation system is designed and implemented based on TM-HSNT.The main contents of this paper are as follows:(1)A new method for calculating user trust is proposed.In EBSN,the number of users to focus only on the calculation of the same group and the number in the same even ts trust tradition,this paper argues that the group or events in the role of different users,to other users of the influence is also different,such as the organizers of other users on the impact of more than ordinary participants.(2)To solve the sparsity of user-activity matrix and active text,a topic model G-wBTM based on group relationship and activity content information is proposed.Active text is usually short text,the traditional topic model is not applicable,and because most users participate in fewer activities,historical data can not support a larger corpus.This paper expands the text through the user group relationship in EBSN.In order to distinguish the importance of text,the model weights the text appropriately,and uses the idea of word-to-word co-occurrence to train the topic model.Finally,the model is recommended according to the similarity between user and activity topic.Compared with the traditional topic model,the accuracy of recommendation has been significantly improved.(3)In order to solve the problems of cold start and insufficient coverage of G-wBTM model recommendation,a topic model TM-HSNT based on heterogeneous social network relations and active text information is proposed.On the basis of the former two research points,a new topic model is constructed,which extends the text by using historical activity information of users with high trust with users,so that recommendation is no longer restricted to groups,and finally recommendation activities for users are based on topic similarity.Compared with traditional recommendation model,activity recommendation based on TM-HSNT improves recommendation accuracy and normalized cumulative loss gain.(4)TM-HSNT model design and implementation of recommendation system based on events.The system includes a user module,information display and events recommended three modules,to achieve the extraction,storage and display features of the data based on the three research points,and through the message queue technology to reduce the response time of the system,finally through the test proved the practicability and reliability of the system.