“…Also, click counts on learning materials (Wolff & Zdrahal, 2012;Wolff, Zdrahal, Herrmannova, & Knoth, 2014), the quantity and quality of forum posts, as well as students' social network positions in forums (Romero, Lopez, Luna, & Ventura, 2013) are also influential features of predictive models. Nevertheless, in a more pervasive perspective, Zhou et al (2018) collected all the web access logs of students in a local university to train predictive models, and found that at-risk students tended to spend more time visiting entertainment-related sites than educational sites. The fact that both types of data are useful naturally incurs the practice of combining the power of both.…”