A Research Of Demographic Information Based On Smartphone’s Streams And Sensor Data

Demographic information, like people’s gender, age, income, education, and religion etc., has great significance in personalized services area, such as targeted advertising and recommender systems. However, lots of users treat these information as their private treasures and reluctant to share them with the Service Provider. To solve this problem, we propose a novel approach to predict the users’ demographic information inadvertently, by using smartphone’s streams and sensor data.Comparing to the computer, smartphone is usually possessed by only one user but not users. Besides, users can take their smartphones in any time because of their mobility. So demographic prediction based on smartphone rather than computer has become a new direction. Additionally, with the popularity of smartphone in the general population, almost every person has their own smartphone. Furthermore, the application developer, who is also called the service provider, can simply get a lot of smartphone usage data through the Application Programming Interface, such as the usage of smartphone applications, the data streams and sensor data. So on the basis of these characteristics, we can get users’ demographic information exactly.In our implement, we analyse the data collected from the smartphone at the first. Then we find that different demographic user has different kinds of smartphone data. So we extract features which we have defined in advance from the stream data and sensor data. And we transform these features to an adjacent matrix R. In the next step, we match the feature matrix with the users and predict the demographic information by using the classifier whose input is the matrix R.In this paper, we design a demographic information prediction system. The system include two sides: the Android Client which is used to collect users’ smartphone data and the Server which uses the smartphone data to predict users’ demographic information and post personalized services to Android Client on the result of users’ demographic prediction.In the experiment, we collect users’ smartphone data from real world and use it to verify our approach. The result shows that the approach based on data streams can achieve 86.50% in Accuracy and 86.43% in Macro F1, and the approach based on acceleration data can achieve 84.64% in both Accuracy and Macro F1.