User Profiles For Personalized Information Recommendation By Authorities In SNS

After several years of vigorous development, the SNS has attracted a large number of users to participate. Users from all levels of society are linked together by releasing,forwarding and commenting microblogs. This interaction process generates a huge flow of information. It makes everybody to retrieve information easily, but also inevitably pushes the problem of information overload to the foreground. So it is of great value to make personalized recommendation by information filtering.It is difficult to filter the information from SNS which has both massive data and users. One of the difficulties is that the threshold of social network information publishing is extremely low, which causes the phenomenon of it mixed all level qualities of information. At the same time the existing technologies of distinguishing the quality of information are relatively inefficient, so that we can both improve the algorithm of deep learning and experts recommending. Their expertise and experiences are the key to help users to filter information. In order to achieve this goal, this study try to find out certain fields experts for information users by detecting their characteristics.SNS users often have some labels to describe their characteristics, through which the user can quickly and accurately be identified and recommended. However, in the social network due to the custom, privacy protection and other factors, the user's personal tags are often not popular enough. This resulted that matching personal information difficulties for sparse user profiles. In order to solve this problem, according to theory of homogeneity, similar users will like similar content, predicting the characteristics of the user by their close friends could be a possible way. In this paper, an improved algorithm based on SimRank named ASCOS were used to calculate the similarity by social relations of users. Users' tags can be predict by the value of ASCOS.Then, a FRank algorithm based on PageRank is proposed to improve the recognition of a small social circle. The results show that the precision and recall of ASCOS in user tag prediction are improved. In expert forecasting unit, the value of nDGC demonstrated that FRank also has a better performance compared with the baseline method PageRank.Finally, based on the above two results, label prediction and expert identification,personalized recommendation is achieved.This article contains 6 chapters:The 1th chapter introduces the current research status and results of personalized recommendation of social network, and the structure of the article.The 2nd chapter explains the theoretical techniques needed to study the paper,including the method of data acquisition, user characteristic model and common recommendation system.In the 3rd chapter, according to the characteristics of Weibo users, user tags are selected as the basis of the recommendation model. And ASCOS algorithm is the key to calculate user similarity and extend the user tags.The 4th chapter identifies the expert users in the network by FRank algorithm which predicting the label and social relationship network, and completes the recommendation system design.In the 5th chapter, by analyzing the data extracted from Weibo, a notable performance of these algorithm were done, which verified the possibility to use these algorithms in practical appliance.The 6th chapter summarizes the results of the analysis, and potential improvement for future study.