Social Search Based On Complex Network

Since21st century, the development of Internet becomes faster and faster. Socialnetworks such as Facebook and Twitter achieve great development only in several years. Theuser count of Facebook is near one billion. The greater social networks develop the deeperpeople research on them. Facebook, Twitter and Yahoo all provide develop interface fordevelopers and researchers. In our country, social networks especially micro blogs appear asrainfalls since2009. Main portals all give out their micro-blog systems. Tent micro-blogallows users register it by using their qq number. Other micro-blogs allow users use email orrenren-acount to register. At the same time, lots of celebrities register micro blogs by usingtheir true names, which makes micro blogs become rather heated. Many organizations alsoregister micro blogs to publish news. Just in a few years, social network has been standing onthe front edge of the Internet. In the meantime of the development of micro-blog, the researchabout micro-blog goes deeper and deeper. Main portals gradually let out their SDK based ondifferent programming languages and different application platforms.Social network is a kind of complex network. Complex network has communitystructures, so it is with social networks. Therefore, the research of social network could takeaccount of the research of complex network. The main researches are centered in graphsegmentation and grade clustering. Kerneighan-Lin method and spectral bisection method arethe main algorithms in graph segmentation. In grade clustering methods, there aresegmentation method and cohesion method. Among the research of community structures insocial networks, threshold method with edge weight and node attributes and SA-Clustermethod are most famous. Threshold method can detect community structures effectively.While SA-Cluster method could solve the cluster problem of communities in social networksby calculating the random walk distances.At the same time, the research of neighborhood is of great value and meaning. In socialnetworks, the network structure is not unchangeable. Many users are adding friends on and on,but as the number of friends grows, some friends may not be the group the user like. Thus, theun-follow phenomenon appears. There are many reasons for un-follow, such as inactive users,rubbish messages, advertisements, and over-posted users. By taking account of these reasons,the formation of friendship will be more reliable and stable. How to provide effective friends recommendation is another key research point in this paper.The search strategies of complex network can also provide references for social search.The main search strategies in complex network contain breadth first search strategy, highdegree seeking and random walk. Breadth first strategy starts searching in global view, whilehigh degree seeking and random walk start from local, which makes their efficiency better.Among the three kinds of search strategies, random walk is most widely used. In social search,there are fast random walk, supervised random walk and a unified framework using randomwalks using random walk strategy. The unified framework method sufficiently takes accountof structure feature and node feature, and builds augmented graph for link recommendation.According to the benefits and shortcomings of various random walk algorithms in socialsearch, we proposed a random walk algorithm for micro-blog. While applying the randomwalk strategy, the algorithm uses recommendation filter strategy and recommendationordering strategy to further filter and optimize the candidate node list, which makes the resultbetter and more stable.According to the search strategies and community structures, and the building andmaintenance of neighborhood friendship, this paper develops a micro-blog recommendationsystem based on random walk by using Tent-micro-blog SDK. The system uses random walkas its strategy and takes account of neighborhood maintenance to filter the recommendationresult, which makes the final result more stable and reliable. Though the system could providebetter results, still many problems need further research, such as the metrics to measure thesearch results, the judgment of user preferences, the decision of the initial parameters byquestion lists and so forth.