Research On Key Technologies Of Media Data Process For Social Network

Since the benefit of the high efficiency,the low cost,Online Social Network(OSN)is widely accepted and used.Especially with the development of mobile Internet,billions of people can connect with others with the help of intelligent mobile devices,and push the latest state,messages,and pictures,audio and video resources on the OSN.This situation leads to the OSN in bloom,but some problems have been occurred,for example,media recommendation for OSN,the trust computation.How to make full use of the resources and structure of the OSN and serve the customers more comformatably is an interesting job.Considering the OSN is the whole of personal interactive relations which combine users,media resources and social relation network,this paper investigates these problems.At first,considering the millions of users and media resources for OSN and clustering method is the main method for dealing with big data problems,the paper investigates the K-Means initialization problem.Based on the current research results,this paper uses Gaussian kernel density estimation and analyzes the density of data in each dimension,and proposes a new K points selection method considering the centers should be away with each other.After projecting the dimension from high variance to low variance,the final K initial centers are constructed.Based on this,another method which uses the product of the distance between highest density location and the point and its' density is proposed.The experiment results show that two methods can achieve good results.Considering the difficulties of personal labelling for millions songs within OSN,this paper investigates the music genre classification method with hierarchical structure.Conventional flat classification method can not make fully use of the classification capacity of the different features between different genres and can not show the distances and the hierarchical relation between different genres.Based on the analysis of the popular methods for music genre classification,this paper proposed a hierarchical method for genre classification.At first,K-Means clustering method is used to get the core features combining the LDA method and the hierarchial structure is built.Extracting the valuable features,a new hierarchical method is built for genre classification.The experiments show that the proposed method can achieve good results.Considering the difficulties of recommending music for users within OSN,this paper also investigates the content-based music similarity method with relevantcomponent analysis.Currently many content-based music similarity methods pay more attention on the whole distribution featues of the music dataset and then gain related feature dataset,but this paper realizes that these methods can not emphasize the key features which are valuable for music similarity.This paper uses PCA whitening to refine the MFCC dataset,and uses K-Means to cluster the core features of the refined dataset,and then with the help of relevant component analysis,the weight of the key features can be enlarged and the weight of the unrelated features can be reduced.In this way,the results show that the accuracy of the content-based music similarity is improved.Considering the millions of users and the difficultis of finding influential users,this paper investigates the personal influence model in OSN.After considering many models,this paper proposed a new personal influence model which contains global influence and local influence.Global influence considers personal's social network,and analyzes the user's influence based on the followers in the OSN.Local influence considers the clubs that the user joins.After analyzing the user's followers in all the clubs,the user's local influence can be computed.Given proper weight for global influence and local influence,the user's influence in the OSN can be predicted.Finally,considering the interactive relations between users exist the trust relation problem within OSN,this paper investigates the trust model in OSN.Conventional methods use the graph model to compute trust in OSN but there exist many problems.In this paper a new trust model for OSN is proposed which combines the interactive trust and reputation trust.Actually,interactive trust is computed based on the interaction degree between the follower and the followee.If both have interactive records,then they have direct interactive trust.Otherwise,they have indirect interactive trust.Reputation trust contains group reputation trust and follower reputation trust.Group reputation trust describes one's reputation in the OSN because of one's reputation.Follower reputation trust describes the reputation that the K most reputable followers have in the OSN.Combining the two kinds trust with proper weights,the trust model can predict one's trust values in the OSN.The experiments show that the proposed model can achieve good results.