Abstract
While similar documents are, traditionally, found using Natural Language Processing, we observe reference/citation information by authors indicates better insight of similarity. Our system is to retrieve publications from Google Scholar and visualize them as a 2D graph using the citation relation, where the nodes represent the documents while the links represent the citation/reference relation between them. We measure the similarity score between each pair of papers based on both the number of paths and the length of each path. More paths and shorter the lengths higher the similarity score. We compared them with another similarity scores from Scurtu's Document Similarity API [1] that uses Natural Language Processing. We use the average of the similarity scores collected from 15 users as a ground truth to determine how good the scores from two methods are. The result shows that our citation network approach gives better results than the ones by Scurtu's.