Research On Overlapping User Community Detection In Micro-blog Network

As the product of the information age in the twenty-first Century,the emergence of the network makes the productivity developing with each passing day.In the tide of technology change,the social network is unwittingly changing our way of behavior,and the social network is full of every corner of our daily life in the era of big data.There are similarities between online social networks and real social networks,both of which possesses community structure characteristics.However,the structural mechanism of online social network is more complex and there are diverse community types for online social network.Community detection is a data mining technology applied in complex networks,which enables people to understand the characteristics of network structure from the objective angle.Therefore,it is necessary to conduct some research on the community detection in the social network.Community detection can be roughly divided into two groups based on whether a user node can belong to two or more than two communities,i.e.overlapping community detection and non-overlapping community detection.Compared with non-overlapping detection,overlapping community detection is similar with the reality network.Although there are many overlapping community detection methods,few algorithms focus on both users' interests and social behavior preferences.Our research mainly focuses on overlapping community detection and community evaluation metrics in complex networks.The main work and achievements are as follows:1)In order to quantify user's interest information,the concept of average partition distance is proposed in this paper,since tags are always considered as interests' indicator for users and can be applied for community division.More specifically,the tag expansion is performed via the inner and outer correlations between tags.And then the tag discrimination degree is defined based on which the core tag sets can be obtained.Besides,starting from the concept of information theory and distance,we propose the pre-community division algorithm based on the mean partition distance of the core tags and the initial community division is performed.2)We propose a comprehensive division method for social network to balance users' interests and social attributes based on the environment of microblogs and users' characteristics.The 6 dimension structure attribute vector is defined according to the user's following/followee relationship and the Attrirank model is integrated to calculate the dissimilarity of user structures.Then,the comprehensive division dissimilarity is derived by adjusting the weight of mean distance of core tag and user structure dissimilarity.Finally,the subgroup corresponding to the tag with the lowest comprehensive division dissimilarity degree is considered as a new community for one iteration until it meets certain requirements.We evaluate our methods through a series of experiments based on a data set crawled from the open API and the results are analyzed.The impact analysis of different parameters are discussed and we also compare our method with basedlines.The results show that our method is capable of overlapping community detection and has practical significance.