Research Article Classification using Graph Convolutional Neural Network

Abstract: The thesis presents a mechanism for research paper classification based on a relational graph that represents the interrelationships among papers, such as citations, authors, common references, etc. The proposed method is a semi-supervised learning that have used graph convolution neural network in learning the relations. The GCNN captures the spatial relation i.e. neighbors of a node in feature vector creation. The Proposed method makes use of message passing system, where node sends their feature values i.e. word embedding to their neighbors node so that every node can create its own feature vector based on the content of its own article and its neighboring articles.  This method has better performances it tries to predict the class based on the neighboring classes and the content of its neighbor which is common between them. Comparison with the previous methods, the proposed method has performed well with a significant margin. Keywords: GCNN, Article Classification