Research Concept Link Prediction via Graph Convolutional Network

As interdisciplinary research develops, numerous publications are uploaded to the research community. This makes researchers spend a lot of time searching for valuable information. Some methods have been proposed to provide more accurate recommendations, such as domain terminology-based or graph pattern-based methods. However, due to the continuous emergence of new research concepts and new research fields, it is difficult to apply it to practical tasks. To tackle these issues, this paper constructs the research keywords of different publications into a topological network, and then predicts the possibility of connection between different keywords. To analyze the associations between research concepts, graph convolution technology are employed to design method. It could provide domain-independent recommendations for the relevance of research concepts. Experiments demonstrate that the proposed method can predict associations between research concepts more effectively than traditional methods.