Seed‐based information retrieval in networks of research publications: Evaluation of direct citations, bibliographic coupling, co‐citations, and PubMed‐related article score

Abstract In this contribution, we deal with seed‐based information retrieval in networks of research publications. Using systematic reviews as a baseline, and publication data from the NIH Open Citation Collection, we compare the performance of the three citation‐based approaches direct citation, co‐citation, and bibliographic coupling with respect to recall and precision measures. In addition, we include the PubMed‐related article score as well as combined approaches in the comparison. We also provide a fairly comprehensive review of earlier research in which citation relations have been used for information retrieval purposes. The results show an advantage for co‐citation over bibliographic coupling and direct citation. However, combining the three approaches outperforms the exclusive use of co‐citation in the study. The results further indicate, in line with previous research, that combining citation‐based approaches with textual approaches enhances the performance of seed‐based information retrieval. The results from the study may guide approaches combining citation‐based and textual approaches in their choice of citation similarity measures. We suggest that future research use more structured approaches to evaluate methods for seed‐based retrieval of publications, including comparative approaches as well as the elaboration of common data sets and baselines for evaluation.