Research On User Group’s Discovery Based On Relation Strength In Social Networks

With the rapid development of Internet, various social networks keep spring up. As aninnovative and convenient mode of making friends, Social network services attract large numbersof users. With various information collected by social networks, more and more users air their views,make friends, and so on. The number of monthly active users on Facebook which is the most famoussocial networking abroad has reached1.1billion. Sina Weibo as a representative of social networkservice in China, has already get more than500million users. Facing a growing number of big dataon social networks, both users and the services providers are desperately trying to solve a problem:How to find people that has common hobbies and views to interact. User group’s discovery comesout for this purpose. Its goals is to find out groups of people with similar interests through miningthe graph of users relationship on social network services, and then supporting advertising, marking,friend recommendation and other practical applications.Traditional methods of user group’s discovery are based on the original graph of relationshipbetween users on social network service. These methods treat users as vertex in graph, treatrelationship of users as edges in graph, and then obtain the user group’s clusters by clusteringanalysis. These traditional methods do not take into account the sparseness of the user relationship,and the difference between relationship of social network and of reality. In the process of findinguser groups, this paper on one hand considers the overall distribution of users’ similar informationin various themes, on the other hand takes into account the influence of the difference in themes’popularity on user relationships. Combining the two aspects above, this paper comes up a model tocalculate the strength of user relationship. This model expands relationships by the characteristicsof social networks, and with the result set of the model this paper implements the user group’sdiscovery by using cluster analysis. The works including:1) At first this paper introduced the relevant technologies, including the basic theory of socialnetwork, computational methods on strength of users’ relationship, the MapReduce programmingmodel and the basic idea of locality sensitive hashing.2) Then this paper presented a method to calculate the strength of users’ relationship byconstructing users’ characteristic co-occurrence vectors. This method combines the diversity indexand weight frequencies, common calculate the strength from two mutually independent perspective. 3) Facing the challenges of the scale of data on social networks, this paper implement theabove calculation through the MapReduce programming model, and then implement the usergroup\s discovery by using the idea of the locality sensitive hash and the features of MapReducewith user relationship strength calculation results.4) By experimenting on the data that obtained through the open ports provided by socialnetwork service Last.fm, this paper estimates the parameters of the model. The results prove thefeasibility and practicality of the strength calculation and user group’s discovery in terms of bothperformance and reliability analysis.