Recommending Relevant Research Papers based on Citation Network Embedding

The work is devoted to the task of recommendation of relevant academic papers based on citation network. Papers recommendation problem is considered as link prediction on citation network. Network embedding based methods and text-based methods are analyzed in terms of papers recommendation task and the performance of the models is evaluated on citation dataset. Among network representation approaches, structural and text-attributed network embedding are considered. Text-based methods include text-similarity and classification model on text embeddings. We showed that the methods based on network embedding are more efficient, but network embeddings are sensitive to the size of training data. Text-similarity methods demonstrate lower performance for citation links prediction, but text embeddings are independent of the incomplete network structure and in general give higher recall.