Multiple Subscriber-Identity-Module Detection Using Social Network Analysis Techniques

Some Telco customers have multiple "Subscriber-Identity-Modules" or SIM cards across competing mobile operators or within the same operator. Multi-SIM detection is beneficial to telecommunications operators in tuning campaigns, acquiring usage from competitors, and retaining customers. The goal was to develop and materialize the techniques required to build a model for detecting such customers, and link between the dials of a single customer within the same operator and/or across different operators. The model was based on Social Network Analysis (SNA), which preliminary necessitated representing the social network graph atop the data warehouse that is continuously fed by switches and charging/billing systems, having customers as nodes and their interactions as links. Centrality metrics were calculated for the nodes along with behavioral attributes for the links, to facilitate for equivalence analysis between each pair of nodes. Its steps typically included detecting the common block of nodes touched by each pair of nodes with respect to the degree centrality of both nodes, and analyzing the way each node in the pair interacts with each node in the common block basing on link attributes. False detection is reduced by calculating the diversity of social circles found in the common block between a potential pair, and counting the sub-edges between both nodes in the pair. Resulting metrics are weighted both manually and using linear regression, and then final scores are used to cluster potential pairs into Multi-SIM probability deciles. The SNA-based approach ensured linking to other dials of a single customer even when belonging to a competitor, and market research analysis proved an arbitrary accuracy level reaching almost 90% when targeting the highest 15% ranking customers. Multiple model verification pinpointed a sweet spot that provides an accuracy level of 84% at Multi-SIM base penetration of 76%. The in-database implementation of metrics ensured a certain level of agility.