With the development of Social Media, web mining analysis has been regarded as one of hot research topics. Traditional single task regression builds models for each task, which ignores the sharing information among tasks in the occasion of multi-variable prediction. Therefore, this paper used multi-task regression mining method, and managed to analyze the pattern between user’s personality and network behavior. This study collected a sample set of 335 RenRen users and 563 Weibo users through online test invitation. Using multi-task regression, the final prediction accuracy is 87％ or more. The result means that multi-task regression works better then single task regression for multi-variable modeling.