Graph convolutional networks in language and vision: A survey

Graph convolutional networks (GCNs) have a strong ability to learn graph representation and have achieved good performance in a range of applications, including social relationship analysis, biological information processing, natural language processing (NLP), computer vision (CV), and so on. In recent years, the application of GCNs in natural language processing and computer vision has attracted substantial interest from researchers, as a result of which many studies based on GCNs have emerged in the fields of natural language processing and computer vision. However, to the best of our knowledge, a comprehensive survey of GCN application in natural language processing and computer vision has not yet been conducted. Accordingly, this survey presents a comprehensive review of the principles of GCNs and its applications in these two fields. First, we summarize the principles of the two types of GCNs, namely spatial methods and spectral methods. Then we divide GCN applications into two categories: natural language processing and computer vision. Subsequently, we present multiple applications from each category in detail. Finally, we outline the limitations of GCNs and discuss possible future research directions. •This survey begins by introducing graph convolutional networks (GCNs), then outlines the two types of GCNs (spectral type and spatial type) and introduces the principles of each type in detail.•This survey goes on to introduce the applications of GCNs in natural language processing and computer vision. At the end of the survey, some shortcomings of the existing GCN framework are highlighted, and some possible improvements are proposed.