Abstract: 【Purpose/significance】There are a large number of excellent papers in the scientific literature that have not been found.Iden⁃tifying and making use of these excellent papers have important practical significance at present【. Method/process】In this study,we use the 1990-2010 original and citation literature data in the field of artificial intelligence from the Web of Science database to construct the original paper-citation feature vector space,and use the decision tree and logistic regression for model training and testing【. Result/conclusion】The result shows that the indicator of " citations during five years after publication" can significantly improve the recogni⁃tion effect of decision trees and logistic regression,making the accuracy of the two models reach 84% and 89% respectively,and the in⁃
crease rate reached more than 20%.The recognition effect of the logistic regression is always better than that of the decision tree.By ad⁃justing the hyperparameters of the two models,the model can obtain a better recognition effect.In addition,early scientific research in the field of artificial intelligence is still in the stage of small team collaboration,and the degree of funding and open access to this field literature is low【. Innovation/limitation】We innovatively introduce machine learning methods to realize the recognition models of “hid⁃den treasures”among massive literature.However,we need apply these recognition models into more disciplines.