CiteDEK: A hybrid EMD-kNN-DTW model for classification of paper citation trajectories

“

A broad debate goes on in literature regarding the number of distinct trajectories. The number varies from a minimum of 2 to 6 classes. Different classification methods capture different groups. Some identified classes are sub-classes, so studying them separately is insignificant. Few works precisely study only specialised trajectories-- Hot Papers (HP) (immediate citations after publication without lasting impact) and Sleeping Beauties (SB) (citations long after a prolonged time of publication). Besides, trajectories with similar trends are named and studied under different classes. For example, HPs are studied separately as early rise-rapid decline, flashes-in-the-pan, sprinters, transient-knowledge-claim, MonDec class, etc. Further, SBs are studied independently as delayed documents, peaklate class, etc. Initial findings suggest that the {\em CiteDEK} model classifies raw citation trajectories of scientific publications with 76% accuracy, which is significant. To the best of our knowledge, it is the first attempt to classify trajectories without explicitly defining or extracting subjective thresholds or parameters. However, one drawback of the 1-kNN-DTW algorithm is that it performs poorly on computationally expensive data sets. In the future, we aim to test the model on larger data sets and improve accuracy using deep learning-based techniques.

”