Many systems can be represented as networks or graph collections of nodes joined by edges. The social structures in these networks can be investigated using graph theory through a process called social network analysis (SNA). In this paper, networks and SNA concepts were applied using Telecom data such as call detail records (CDRs) and customers data to model our social network and to construct a weighed graph in which each relation carries a different weight, representing how close two subscribers are to each other. In addition, SNA is used to explore the Telecom network and calculate the centrality measures, which help determine the node importance in the network. Depending on centrality measures as well as influence capability of node measure, the influencers in network were detected and targeted by marketing campaigns resulting in 30% raise in growth rate of mobile traffic compared with traditional ways. Finding Multi-SIM subscribers within the same operator or across different operators presents another important concern to Telecom companies because it allows to improve campaigns and churn prediction models. Social network similarity measures and social behavioral measures between nodes were calculated in the Telecom network to detect these Multi-SIM subscribes and 85% accuracy result was achieved for subscribes from different operators and 92% for subscribes from the same operator. The paper is based on a real dataset of 3 months CDRs and customer data provided by a local Telecom operator. This dataset is used to build a network with more than 16 million nodes and more than 300 million edges on a big data platform.

中文翻译：

---

电信数据中的社交网络分析

许多系统可以表示为通过边连接的节点的网络或图形集合。可以使用图论通过称为社交网络分析（SNA）的过程来研究这些网络中的社会结构。在本文中，通过使用电信数据（例如呼叫详细记录（CDR）和客户数据）来应用网络和SNA概念来对我们的社交网络进行建模，并构建一个权重图，其中每个关系具有不同的权重，代表着两个订户之间的距离对彼此。此外，SNA还用于探索电信网络并计算集中度度量，这有助于确定网络中的节点重要性。根据中心度量以及节点度量的影响能力，通过市场营销活动发现并定位网络中的影响者，与传统方式相比，移动流量的增长率提高了30％。在同一家运营商或不同运营商中寻找Multi-SIM用户是电信公司的另一个重要问题，因为它可以改善活动和客户流失预测模型。通过在电信网络中计算节点之间的社交网络相似性度量和社交行为度量，以检测这些Multi-SIM订阅，不同运营商的订阅的准确性达到85％，同一运营商的订阅的准确性达到92％。本文基于3个月CDR的真实数据集和当地电信运营商提供的客户数据。