Research On Ensemble Recommend Algorithm Based On Social Network

With the rapid development of Internet technology, the information age has been unavoidably walk into our lives. When our lives everywhere need information and we are helpless due to face to growing surge of information day by day, the information overload problem would arise. Under the above backgroup, the recommendation system came into being. On the other hand, the rapid rise of social networks would have promoted to the great development of information exchange between our communication, we would have generated more information by communicating with others in social networks. However, with our entering the era of social nerwork, user’s role has made a fundamental change. At this time, the user would not only be receive information, but also manufacture information and play the role of disseminating information. Thus, the information overload problem in social networks is particularly serious. So, research on recommend algorithm based on social network is moreimminent than other things.As we all know, there would be still many problems to be solved. And the recommendation models on the social network would have brought many challenges for the traditional methods. This paper would face to the complexity of social networks, in order to solve some new problems and old problems which exist in the traditional methods:Multiple Heterogeneous on data, cold start problem, dynamic problems, user-level problems,sequence mode and other issues. In this paper, we combinate recommendation algorithm based on matrix factorization with linear regression and bilinear regression model to design an ensemble recommendation model. And we would analysis 6 potehntial properties impacting on recommendation system on the social network:social activities, user profiles, contextual information, hierarchical information, dynamic information, the user sequence mode. We would design proper models to caputure every potehntial properties based on the characteristics of these six factors. Then,we would design the final ensemble recommation model to caputure all potehntial properties. Finally, we choose data of the microbog as the data set on the experiment, and run every the ensemble recommendation models to analyze and compare the experimental results in order to prove our hypothesis and our ensemble recommendation models could solve the issues on the above, the new model is also recommended certain extent, increase the recommended accuracy.