Study Of Mining Social Network Based On Call Records

Since telephone was invented in the19th century, People develop the mobile phone forbetter communication in the20th century. With the rise of the Internet in the21st century,mobile Internet was more widely used now.As a communication technology, telecommunica--tions is constantly changing the way people live.It also accumulated a lot of data at the sametime. We are in an era of big data. The traditional way of the price adjustment, equipmentupgrades can not be able to create significant value for the telecommunications industry.So,many telecom companies are transferring to the field of competition on user-data and callrecords. How to mine valuable information which we can’t see before from the existing datahas become an hot topic.As a topic of data mining, association discovery has been extensively studied for years.Many scholars in this field made a lot of new theories and algorithms, such as Kernighan-Linalgorithm, Laplace algorithm, GN algorithm and etc. Some of the algorithms showed goodmining results in the actual network. However, due to algorithm efficiency or limitations onthe structure, the algorithms can not be widely used in all fields. Most of the algorithms usedto find balance between the accuracy and efficiency.Telecom users constitute a great social network, but the network is relatively sparse.Traditional algorithms greatly influenced by discrete points. It’s important to give an algorithmwhich is suitable for mining social network relationships from telecom data on the basis ofexisting theoretical knowledge. This article comes up with a new algorithm which based onsocial triangle theory. Firstly, it finds all the triangular-relationship user groups as initialcommunity in network, then these communities expansion according to nodes-similarity. Forthose points which have not been assigned to any society are treated as discrete points.It can excludes the impact brought by the discrete points effectively. In article, I willapplied the algorithm to the telecommunications data for testing and compared with thetraditional algorithm. In the end, I will analysis the existing problems of the algorithm andgive the improvement directions needed for further research.