Citation Patterns Analysis And The Early Identification Model Construction Of Transformative Research

Transformative research(TR)involves research that radically change our current understanding or leads to the creation of a new paradigm or field of science,provides pathways to new frontiers.Since the 21st century,major developed countries in Europe and the U.S.have vigorously strengthened the exploration and support for TR in order to maintain their global technological leadership and international competitiveness.The Chinese government,in recent years,has also laid high emphasis on TR.It has clearly pointed out that "to increase the support for non-consensus and transformative innovation research effectively,and encourage the research that challenge authority and tradition,that may reshape important scientific or engineering concepts,leads to the creation of a new paradigm or field of science." In this context,the early identification of TR has important strategic significance for seizing the commanding heights of international S&T competition.However,there still remains a roadblock for the early identification and cultivation of TR:Firstly,there is still lack a unified understanding on the concepts and characteristics of TR among researchers.Secondly,the theoretical model and characteristic index are still immature,and a widely accepted early identification model is still missing.In view of the current situation,we firstly reviewed the concepts and essences of TR,and constructed a generation and diffusion concept model of TR based on the structural hole theory,theory of scientific revolution,and innovation diffusion theory in view of citations.In addition,the paper deconstructed the generation and diffusion process of TR from the viewpoint of TR's references and citing articles,and build a generation model and diffusion model of TR.These models revealed the knowledge combination characteristics and the knowledge diffusion characteristics from the time,disciplinary,and impact dimension.The paper also constituted an early identification model of TR based on commonly used indicator selection,pattern analysis,and model fitting process.Then,the paper analyzed the citation pattern of Nobel Prize articles,from the time,discipline,and impact dimension based on the theoretical model.Results suggested that aging characteristics,disciplinary similarity characteristics,and citation impact characteristics were the typical reference characteristics of TR.From the perspective of references,the typical characteristics of TR's references were mostly derived from the same or similar disciplines,the aging speed is faster,and have relatively higher academic influence.On the other hand,the typical characteristics of TR's citing articles were mostly multi-spread to distant disciplines,the early spread speed was faster,and its impact was higher.Thirdly,entropy weight method,factor analysis method,and Logistic regression model were applied to construct the early identification model of TR based on the typical citation characteristics.The identification effect of different models was further verified and compared based on the research areas which had won the Nobel Prize.After the data consistency,and correlation analysis of the typical citation characteristics indicators of TR,the paper selected one of the best early identification models to detect TR according to the ranking of Noble Prizewinning paper(NPs),and analyzed the accuracy rate of this model furtherly.Results show that,firstly,the identification effect of Logistic regression model was the best model among all the models,which can detect NPs the third year after its publication.In addition,the NPs ranks at the very top of all TR.Secondly,the best identification model can pick up 3 NPs among 4 TR.Besides,it will spend 9 up to 19 years for these NPs from the early years after its publication to the time that were awarded Nobel Prize.Finally,we take molecular biology,polymer science,and astrophysics as examples to conduct an empirical study by using the model that trained with the help of machine learning techniques.In addition,this article also conducted an early assessment on the early identification results based on bibliometrics,citation sentiment analysis,and the guidelines for the special declaration of key scientific issues of transformative technology from the government.Results show that these early identification results were all published on high quality journals with funding.However,this early identification results' citations and usage data did not have any correlation.Citation sentiment analysis shows that 87.5%of these early identification results aroused attention from peers with pros and cons,which meets many of TR's characteristics.Finally,50%of the early identification results can be found in the guidelines of the mainly supported items from Chinese government in the 2021.The results suggested that the recognition effect of early identification model of TR was initially verified.