Relationship Circle Mining Based On The Records Of Calls And Messages

With the rapid development of Communication Engineering, mobile phones are moreand more pervasive and the number of telecom users is not easy to increase. There is fiercecompetition between telecommunication operators in developing new customers andkeeping old customers. On the other hand, many mobile phone apps such as QQ and microletter which can used to send short messages and talk to others are developed by big internetcompanies. These bring a big impact to telecommunication operators. So, they must tounderstand their customers deeply and to provide personalization service. Mining therelationship circles between customers through the historic data is a good way to understandtheir customers and this received many attention from big telecommunication operators.Community detecting technology is the way to dig out the community structure fromcomplex network by using related algorithms. For a social network, a community is arelationship circle which is formed by similar users. Many classical community detectingalgorithms had been proposed, such as hierarchical clustering algorithms, spectral analysisand LPA and so on. These algorithms can’t find overlapping communities. Palla proposedthe first overlapping community detecting algorithm in2005and many other overlappingcommunity detecting algorithms had been proposed after that.Most of the existing community detecting algorithms are only based on the structure ofthe complex network and ignore the strength of relationship between two nodes so thatthese algorithms are not very fit for the networks which have weight about edges. For thisproblem, this paper bring the information about the strength of relationship between nodesinto the CPM, then proposed a new algorithms called SR_CPM which can apply to thenetworks which have weight about edges.For the network of telecommunication users, this paper proposed a method to calculatethe strength of relationship between two nodes. This method combined three aspectinformation, the intrinsic properties of user，the records of calls and the records of message.At last, this paper apply the SR_CPM to the data of the telecommunication users network. The result shows that the SR_CPM is better than the CPM under the Q and v evaluationindicators.