Recommending Relevant Research Articles based on Citation Graph and Textual Information

With the ever increasing number of published research articles, it is important to develop new solutions for getting better recommendations than a naive keyword based search. A plethora of methods are proposed for citation recommendation by using either the citation graph that exists between the research articles or the textual information embedded in the research articles. Taking inspiration from recent developments, we propose a novel method by combining the feature vectors obtained from doc2vec and node2vec where the former extracts the text features and the latter extracts the citation network features using Canonical Correlation Analysis. We have elaborated the proposed hybrid approach on the real-world Cora citation network. We observed that the ensemble method outperforms the state-of-the art distributed method representations.