Large-scale analysis of delayed recognition using sleeping beauty and the prince

Abstract

Delayed recognition in which innovative discoveries are re-evaluated after a long period has significant implications for scientific progress. The quantitative method to detect delayed recognition is described as the pair of Sleeping Beauty (SB) and its Prince (PR), where SB refers to citation bursts and its PR triggers SB's awakeness calculated based on their citation history. This research provides the methods to extract valid and large SB-PR pairs from a comprehensive Scopus dataset and analyses how PR discovers SB. We prove that the proposed method can extract long-sleep and large-scale SB and its PR best covers the previous multi-disciplinary pairs, which enables to observe delayed recognition. Besides, we show that the high-impact SB-PR pairs extracted by the proposed method are more likely to be located in the same field. This indicates that a hidden SB that your research can awaken may exist closer than you think. On the other hand, although SB-PR pairs are fat-tailed in Beauty Coefficient and more likely to integrate separate fields compared to ordinary citations, it is not possible to predict which citation leads to awake SB using the rarity of citation. There is no easy way to limit the areas where SB-PR pairs occur or detect it early, suggesting that researchers and administrators need to focus on a variety of areas. This research provides comprehensive knowledge about the development of scientific findings that will be evaluated over time.

Keywords: Bibliometrics; Citation network; Delayed recognition; Interdisciplinary fusion; Prince; Production of knowledge; Sleeping beauty.

Fig. 1

Overview of how a PR discovers an SB in the citation network. Each node represents papers and edges shows their citation. (1, left) As soon as an SB is submitted, it is indistinguishable from ordinary papers in the community without being cited much. (2, middle) When the PR realised the importance of the SB and cited it, it spread to the rest of the scientific community and led to the paper being co-cited. (3, right) A sub-field created by a pair of papers becomes widely recognised, and a new field can emerge from it

Fig. 2

Publication year of papers. The black bar shows the whole number of papers published in the year. The orange bar represents the top5% in citation normalised by the average number of citations of papers submitted in the same field in the same year (Radicchi et al. 2008). The small graph shows the proportion of top papers to the total number of papers submitted in each year, which was stable at around 0.05 for papers in the period when there was sufficient time to obtain citations

Fig. 3

Illustration of the definition of the beauty coefficient B and its awakening year $t_a$. The black curve represents the number of citations $c_t$ received by the paper at age t. The awakening time $t_a\le t_m$ is defined as the age that maximises the distance from the line connceting points ($0,c_0$) and ($t_m,c_{t_m}$). B is calculated as a summation of the ratio of blue lines against each blue dot line from 0 to $t_m$

Fig. 4

Distribution of B for top papers. A red dot line indicates the least score of SBs behind the peak of the majority of papers

Fig. 5

Distribution of citation in each year for all papers. Each coloured line refers to the average number of citation published in dotted point. Although citation inflation increases the citations to papers submitted in recent years, it does not increase the number of citations to older papers

Fig. 6

Yearly citation curve of the lowest B-score in SB papers. The dotted line indicates the year $t_a$ when the paper was discovered

Fig. 7

Sensitivity of clustering in 2020. Every sub-clustering, the maximum size of cluster is reduced by about 1/10, and the number of clusters is increased nearly tenfold

Fig. 8

Cluster size in 2020. The cluster size increases on a log scale in the use of modularity maximisation

Fig. 9

Publication year of SB and PR extracted in 2020. The orange bars show the distribution of submission years for SB, and the blue bars show the distribution of submission years for PR

Fig. 10

Probability of gap year between SB and PR. Length of sleep duration for SB and PR pairs in the dataset divided into 5-year increments. The mean values are 16.0, 19.3, 22.7, 25.9, and 27.9 years, which is about half of the total length of the dataset

Fig. 11

Illustration of the null model. Each node (paper) has a potential of $k_o$ (the number of black dot), which is defined by its 2020 citation, and every after adding new nodes every year, new edges are created with equal probability based on the remaining black dots

Fig. 12

Distribution of B in the dataset and null model. The thin line is the result of log-normal fitting using the maximum likelihood method on each data

Fig. 13

Density distribution of d for inter-disciplinary and intra-disciplinary at the time of their citation

Fig. 14

Density distribution of d for random pairs and SB–PR pairs