Personalized Dissatisfied Users Prediction in Mobile Communication Service

Yunong Chen, Yuying Lin, Bojian Zhang, Dongming Zhao, Haiwei Zhang, Yanlong Wen

Abstract: In the mobile communication industry, users are always concerned with the services provided by the operators. Users who are dissatisfied with the services will probably change their mobile network operators. Therefore, mobile network operators desire to predict whether users will be dissatisfied with the services by analyzing users’ events. Then they stand a chance to timely remedy the services for potential dissatisfied users. Though many existing classification methods are available, they cannot leverage user attribute information well in this task. To address the problem, we propose a Personalized Attention-based Long Short-Term Memory (PA-LSTM) model, consisting of events feature extraction module, user feature extraction module, and personalized prediction module. PA-LSTM makes personalized predictions of dissatisfied users based on both user events and user attributes. Furthermore, PA-LSTM considers the satisfaction tendency of different user groups. Extensive experiments on the industry dataset show that our model performs better than other solutions, verifying the effectiveness of our model.