Relevant Techniques Of User Influence And Personalized Ranking In Social Networking Service

Social networking service is becoming an important channel and carrier for information spreading and keeping up human social relations.Studies about theory and core technology of social networking service have practical application value for social development and business service application.User influence analysis,one of the key contents of social networking service analysis,has been widely used in many areas,such as recommended system,social advertising,link prediction,real-time event detection,etc.Featured by instant release,real-time transmission and convenience,microblog has gradually stepped into the rank of the most popular social network platforms.User's influence,which is of great importance to optimize and motivate social information transmission,plays a basic as well as important role in microblog social network.Microblog users' influence,a basic feature of microblog,has attracted numerous scholars to carry on researches as it faces with large-scale microblog user group.Personalized ranking is an importance ranking for network nodes which based on network linking structure and user preference.Personalized ranking technology in social network has vital research significance in link spam detection,friend recommendation,precision marketing,community detection,etc.Meanwhile,online social media in real life has large network structure and are evolving rapidly such as Facebook,Twitter and so on.Therefore,facing with personalized ranking technology featuring instantaneity,we should design an scalable algorithm which can update the importance ranking dynamically in order to cope with the envolving networks effectively.Aiming at analyzing and researching user influence in social network service and personalized ranking,this paper has the following findings and contributions.(1)In order to solve the unreasonable problem of distributing the PageRank value to all its outdegree nodes in average of PageRank algorithm in its iterative process,this paper analyzes the difference between microblog users' quality and then introduce the concept of users' relative quality.Meanwhile,this paper also analyzes such aspects as comment rate,transpond rate and verification of microblog users.(2)Confronting large scale user data,reasonable and effective parallel processing appears very important.Combining MapReduce programming model,we can design a new user influence ranking algorithm based on PageRank.The experiment result under Hadoop platform shows that with great scalability,the proposed algorithm can effectively combine the relationship and behavior features of microblog users and thus reflect the actual user's influence in a more convincing and sufficient way.(3)This paper first analyzes the local update method of personalized PageRank.For the sake of analyzing complexity of algorithm under dynamic network structure,we introduce the random permutation edge arrival model.Based on the local update method of personalized PageRank,we first present the PriorityPush algorithm which has added the priority queue and negative residuals.Then,we design the DynamicPriorityPush algorithm for tracking the dynamic network structure changes.Based on the random permutation edge arrival model,this paper presents detailed complexity analysis of this algorithm and verifies effectiveness and accuracy of this algorithm through experiment.(4)The experiment result shows that the proposed method can track about 400 edges' changes within Is.In particular,for WikiTalk dataset,the proposed method can update personalized PageRank value within 390us for single edge deletion.The experiment result also validates that the running time of the proposed method has a linear relation with the number of edges changed.For any k changes,the average running time of our method is O(d/?+k+k/(n?)).For the every changed edge,the running time of our method is O(1/?)in amortized analysis.Meanwhile,this dynamic algorithm has better operating efficiency than the other two methods while the accuracy can be guaranteed.Compared with PriorityPush from scratch method,the proposed method is 23-114 times faster and more than 455 times faster than Monte Carlo method at most.