Identifying potential sleeping beauties based on dynamic time warping algorithm and citation curve benchmarking

Abstract

Sleeping beauty (SB) is recognised as delayed, highly cited, or high-impact literature. The precise and efficient identification of potential sleeping beauties from the massive literature can maximise their value in science and technology development. Therefore, in this study, a new time-series similarity method, called the dynamic time warping (DTW) algorithm, was designed to efficiently identify sleeping beauties. First, the top 1 % of highly cited publications (5423 publications) in the field of Artificial Intelligence (AI) between 1990 and 2010 were identified based on data collected from the Web of Science database. Then, the DTW algorithm was designed and implemented to identify potential sleeping beauties based on the benchmarking sleeping beauty citation curve. Subsequently, the DTW method was combined with three indicators defined by Van Raan (2004) to design the DTW* method. Finally, the newly designed DTW and DTW* methods were used alongside quadratic function fitting (QFF), beauty coefficient (B), and modified beauty coefficient (Bcp) to identify sleeping beauties and evaluate the effect of the DTW algorithm. Among the findings: (1) The DTW algorithm can quickly and effectively identify potential sleeping beauties with help from the citation trajectory of different benchmarking sleeping beauties. These benchmarking sleeping beauties are identified by Raan's criteria from physics, as well as the B and Bcp index from the AI field, indicating the robustness of the DTW method, which is less reliant on methods and disciplinary factors of selecting benchmarking sleeping beauties. The DTW algorithm can automatically and accurately identify different types of highly influential publications, including sleeping beauties and highly cited publications, based on publication citation trajectories, further indicating robustness and application prospects. (2) The DTW* method improves the DTW recognition accuracy using the sleeping time defined by Van Raan, to accurately identify standardised sleeping beauties conforming to the standardised citation curve of benchmarking sleeping beauty, thereby ensuring exclusion of false sleeping beauties with extremely short sleeping time. (3) Using the DTW* method, 39 sleeping beauties with sleeping time greater than five years were identified, among which, 38 met Raan's three criteria with an accuracy of 97 %. Of the 39 sleeping beauties, 62 % were conference publications, suggesting that conference publications have an extremely high probability of becoming a sleeping beauty in the AI field. In content analysis, the 39 sleeping beauties were associated with innovative algorithms, methods, and related applications. (4) The DTW algorithm can be extended to another significant different field as ‘Social Sciences, Interdisciplinary’ category for sleeping beauty identification, further verifying the effectiveness and robustness of the DTW approach.