Research Of Publish-subscribe Mechanism Based On Text Classification In Social Network Sites

As SNS(Social Network Sites, SNS for short) is more and more popular, the waypeople get information changes a lot. In earlier days, people read the news recommendedby editors in web portals, but now they prefer to enjoy the information shared by theirfriends on SNS. The keep-growing information and increasingly diversified user grouprise new challenges to community owners: how to provide various accesses for users topublish new information on SNS whenever and wherever they are; how to find theinteresting ones out of the mass disorder news in SNS and present them to the users.Taking the requirements of an actually running Social Network Site intoconsideration, we develops a publish-subscribe framework which can be easilytransplanted into any other similar sites:1) Through deep understanding of differentcommunication methods, supporting various accesses to the community, including butnot limited to: short messages of cell phone, PC clients, web page and so on, enablingusers to publish and receive the trends in SNS whenever and wherever they are.2) Byintroducing publish-subscribe mechanism into the SNS to provide users with the functionof topic-based subscription. Combine the users’ feedback with traditional kNN textclassification algorithm to improve the accuracy of text auto classification and make theresult of classification more compatible with the real status of certain SNS.3) Coming upwith a dynamic topic cluster algorithm to help users get information published underrelated topics with his subscribed ones. By the work done above, the framework can helpfind the be-wanted information out of the mass information in SNS and present them tocorresponding users according to their interests.After actual implementation and experiments, accesses provided by the frameworkcan work smoothly and normally. The publish-subscribe mechanism based onauto-classification of text and dynamic topic clustering can work at an accuracy over80 per cents in a relatively short delay and satisfy the needs of actual implementation.