Age prediction based on the metadata, content and stylistic features of microblogs

作者：Guo Lei; Ma Jun^*; Chen Zhumin

来源：Journal of Computational Information Systems, 2012, 8(16): 6733-6740.

• Age-related
• Hybrid features
• Mean absolute error
• Online social networks
• Personalized service
• Prediction-based
• Predictive abilities
• Regression model
• Regression problem
• Sina-weibo
• Support vector regression (SVR)
• SVR
• Targeted advertising
• Topic model

Currently, online social networks become the platforms for collaboration and communication. It helps make it possible to provide personalized services, e.g., targeted advertising, recommendation and marketing. All these applications are age-related. In this paper, we explore the age prediction issue in microblogs. Contrary to previous research, we treat this task as a regression problem, where we use Support Vector Regression (SVR) method to examine the contributions of three kinds of features, i.e., metadata, content and stylistic. To find the best way for predicting, we train SVR models with one kind of features respectively as well as their hybrid features. The experimental results show that the model using stylistic and content features gains the best correlation (r) of 0.712 and MAE (mean absolute error) with 7.112 years. Besides, we train Support Vector Machine to investigate the predictive ability of these features in classification problems, where we get similar conclusions as shown in regression models.

• 出版日期2012