Paper recommendation using citation proximity in bibliographic coupling

Research paper recommendation has been a hot research area for the last few decades. Thus far, numerous different paper recommendation approaches have been proposed. Some of these include methods based on metadata, content similarity, collaborative filtering, and citation analysis, among others. Citation analysis methods include bibliographic coupling and co-citation analysis. Much research has been done in the area of co-citation analysis. Researchers have also performed experiments using the proximity of in-text citations in co-citation analysis and have found that it improves the accuracy of paper recommendation. In co-citation analysis, the similarity is discovered based on the frequency of co-cited papers in different research papers and those citing papers may belong to different areas. However, when proximity is used to calculate co-citation, the accuracy of recommendations improves significantly. Bibliographic coupling finds bibliographic coupling strength based on the common references between two papers. In bibliographic coupling, a large number of common references of two papers means that they belong to the same area, unlike co-citation analysis, in which there is a possibility that the citing papers may belong to different areas. Based on the observation that with the use of proximity analysis the accuracy in cases of co-citation analysis has improved, this paper investigates if the accuracy of paper recommendation can be further improved by using proximity analysis in bibliographic coupling. This paper proposes an approach that extends the traditional bibliographic coupling by exploiting the proximity of in-text citations of bibliographically coupled articles. The proposed approach takes into account the proximity of in-text citations by clustering the in-text citations using a density-based algorithm called DBSCAN. Experiments on a data set of research papers are presented to show that there is a substantial increase in accuracy of the recommendations produced by DBSCAN based on proximity analysis of in-text citations compared to traditional bibliographic coupling and content-based approaches.