Combining unsupervised and supervised classification for customer value discovery in the telecom industry: a deep learning approach

Customer behaviour analysis in a telecom market is a challenging task in the customer relationship management area. In this paper, we propose a customer behaviour recognition model that combines unsupervised classification and supervised classification methods. First, considering the complexity and uncertainty of consumption behaviour, a hybrid model of K-means clustering, the entropy method and customer portrait analysis is applied to segment customers. Second, the segmentation results are subsequently incorporated into the proposed multi-head self-attention-based nested long short-term memory classifier to evaluate the performance of customer behaviour recognition. Third, the proposed framework is applied to a real case obtained from the China telecom market. The results indicate that our model is significantly superior to other traditional customer behaviour classification models. In addition, medium-value customers will make full use of the mobile traffic packet, and the package utilization rate of high-value groups is lower, which may benefit the precision marketing of telecom companies.