Demographic Prediction Based On Users’ Smartphone Application Log

Demographic information is usually treated as private data( e.g., gender and age), but it has been shown great values in personalized service, advertisement, behaviors study and other fields. A novel approach is proposed to make efficient demographic prediction based on smartphone applications usage.Specifically, the data set is characterized by building a matrix to correlate users with types of topics from the log file of smartphone applications firstly. And secondly, combined with the users’ demographic information, we analyze the preference of different demographic user groups. While providing the baseline prediction with four classical classifiers, we find that the prediction results can achieve stable state with the duration of 8 weeks data. Considering the topic-unbalance problem deeply, we propose an optimization method to further smooth the obtained results with topic neighbors and user neighbors.The evaluation is supplemented by the dataset from real world workload to show advantages of the proposed prediction approach compared with baseline predictions. In particular, the proposed approach can achieve 80.11% and 81.21% of F1-score and Accuracy in gender prediction, respectively. While in dealing with a more challenging multi-class problem, the proposed approach can still achieve good performance, which is more than 71.82% of F1-score and Accuracy in the prediction of age group and more than 64.39% of F1-score and Accuracy in the prediction of phone level.