Feature Mining And Classification Prediction Of ESI Highly Cited Papers

High-quality academic papers play an important role in promoting the progress and development of science,and exploring their typical characteristics is of great significance to predict the future influence and automatic recommendation of papers in advance.Because of the authority and extensive coverage of its database,the ESI index of highly cited papers put forward by Clarivate is often used to measure the scientific influence of individuals,institutions and countries,and these ESI papers are often regarded as high-level and high-quality papers.This paper will take the ESI papers in the field of physics as the research object,by comparing the characteristic differences between these papers and ESI papers to reveal the typical characteristics of highly cited papers,and use these characteristics to predict whether a paper can become an ESI highly cited paper in the future.This paper takes the relevant paper data of Web of Science,APS and altmetric.com as empirical data,and selects the four-dimensional measurement indicators of paper originality,reference list features,alternative metrological features and interdisciplinary characteristics.Firstly,this paper makes an annual statistical comparative analysis of the four-dimensional indicators of ESI papers and non-ESI papers published in PRB,PRD,PRX and PRL journals from 2011 to 2018.Then use a variety of machine learning algorithms to classify and predict ESI papers,and use a variety of indicators to evaluate the model,and finally rank the importance of four-dimensional measurement indicators according to the model with the best prediction effect.The results show that compared with non-ESI papers,ESI papers have higher originality and social attention,and tend to quote "recent hot" papers,but there is no significant difference in interdisciplinary characteristics between ESI papers and non-ESI papers.In addition,by analyzing the correlation between the four-dimensional measurement indicators of the paper,it is found that there is a strong correlation between the originality of the paper and the characteristics of the reference list.In the prediction of ESI papers,the stepwise logical regression model is the best among the four classified prediction models.Finally,this paper also finds that originality is the primary factor that affects the paper to become an ESI paper.