Issue-based information system (IBIS) is a classic argumentation-based approach to solve wicked problems. One of its important problems is how to classify labels for its nodes such as ideas and issues from argumentation documents, which usually causes a huge burden. In this paper, we propose a graph convolutional networks (GCN)-based method that can automatically classify the node labels by efficiently utilizing the relationship of the nodes. Specifically, we consider two kinds of IBIS structures: strict-IBIS and soft-IBIS which are distinguished by whether the argumentation structures strictly follow IBIS definition. We then perform evaluations on a real English discussion dataset to confirm the effectiveness of our proposed method.