Identifying Loners from Their Project Collaboration Records - A Graph-Based Approach

Identification of lonely students is important because loneliness may lead to sickness, depression, and even suicide for college students. Loneliness scales are the general instruments used to identify loners, but it usually fails when loners try to conceal their real conditions in the questionnaires. In this paper, we propose a framework for the identification of loners based on their project collaboration records, a relatively more objective data source than student's self-reports. Considering that collaborative relationships among students are highly informative for the identification of loners, we employ Graph Neural Networks to model the complex patterns of student interactions. Furthermore, we propose a Graph-based Over-sampling Technique (GOT) to address the class-imbalanced problem for graph-structured data. Experiments on a real-world dataset show that our proposed method can identify loners with high accuracy.