Massive open online courses (MOOCs) have become a way of online learning across the world in the past few years. However, the extremely high dropout rate has brought many challenges to the development of online learning. Most of current machine learning, deep learning, and ensemble learning methods have deficiency in learning effective features for Early Dropout Prediction (EDP). In this paper, we have proposed an early dropout prediction method, and developed a system for the personalized education guidance of student and teachers. Attention-based document representation as a vector (A-Doc2vec) is proposed to learn sequence features of student behavior, and heterogeneous weighted soft voting classification model is proposed to improve the dropout prediction accuracy. This system can reduce the dropout of MOOC Courses in advance.