Research And Implementation Of Group Recommendation Algorithm In Active Social Network

With the rapid development of Internet technology,people's communication methods have also changed,which has prompted the emergence of event-based heterogeneous social networks(H-EBSN,Heterogeneous Event-based Social Network).In H-EBSN,users can communicate and organize activities online and offline,which has also brought about the rapid growth of information,leading to information overload,that is,finding interesting activities is becoming more and more difficult.Personalized recommendation is an effective way to solve this problem.In the existing recommendation systems,most of them are recommended by individuals.However,in reality,the recommended objects do not only exist as individuals.There are more and more groups.Users often participate in activities in the form of groups.The individual-based recommendation method is no longer applicable,so it is necessary to recommend activities for groups of multiple users.Compared with the traditional recommendation field,group recommendation also has a serious cold-start problem.In addition,the diversity of user preferences makes it difficult to find activities that meet the preferences of all members,which poses a huge challenge to group recommendation.Therefore,this paper studies the group recommendation of active social networks,and forms a multi-task learning algorithm based on heterogeneous network embedding and convolutional neural network(CNN Multi-task Learning for Heterogeneous Network Recommendation,CMHR).The group recommendation of H-EBSN has been studied to provide recommendation services for groups in H-EBSN.The main research points of this article are as follows:(1)Constructing H-EBSN,obtaining and optimizing the link path,using the meta-path-based probabilistic random walk strategy and down-sampling to generate the heterogeneous network entity embedding vector,and achieved better prediction results.(2)A preference aggregation strategy is proposed,which aggregates weights based on the similarity between users and between users and groups.On the basis of ensuring the accuracy of recommendations,it better meets the preferences of each member of the group.(3)The cross-sharing unit is designed based on the convolutional neural network,and information is converted and complemented between the recommendation system and the knowledge graph system,and the information between the two can be used to better solve the problem of data sparseness.(4)A group recommendation system was designed and implemented,and the above algorithm was introduced for group recommendation to meet the needs of users.