Ranking Cellular Internet Card User VIP Values Based on Graph Clustering with Usage Data

Internet card (IC) is a new business model in telecom operators, which has attracted numerous users. However, the high user churn rate and the lack of IC user VIP value classification methods make the development of IC business encounter a bottleneck. In this paper, we conduct an in-depth study of IC user VIP value characterization and ranking by utilizing a large-scale dataset collected from a provincial cellular operator of China. We combine the knowledge in the field of traditional business user value management and new feature mining, and devise a data-driven system named REAL, i.e., Ranking cellular InternEt cArd user VIP vaLues, which consists of a dataset collection component, a feature extraction component, a graph clustering model (GCW) and a result output component. In REAL, consumption features, dependency features, and penalty features are extracted, and GCW model is devised to learn the common relationship among different IC users. Finally, a weight-based approach is designed to rank each user and output their VIP grades according to a threshold. Extensive experiments corroborate the efficacy of REAL in ranking IC users’ VIP values.