Evaluation Of Innovation Ability Of Scientific Research Institutions Based On Data Mining Methods

Economic globalization has gradually changed the mode of economic development from resource-dependent to innovation-driven,and international competition is increasingly reflected in high-tech and innovation capabilities.Therefore,KDD CUP 2016 proposes a scheme that uses public data to ranking research institutions based on the amount of articles accepted in 8 high-impact international conferences.This paper has obtained article information published by741 affiliations from MAG dataset and it was used as a basis for research to predict the influence of the institution.Specifically include:1)Data preprocessing.From the MAG dataset,information of articles published by 741 institutions over the years was obtained,preprocessing such as denoising and information matching were performed,a brief summary of evaluation criteria and evaluation methods of common innovation capabilities was made.2)Evaluation of innovation ability of scientific research institutions based on network features.This paper builds a co-authority network based on articles published jointly by different institutions.And uses important node mining algorithms to find out more influential institutions in the network to predict the future influence of the institution.At the same time,the features that can represent the influence of each node in the network are taken as the network center feature of the institution.3)Evaluation of innovation ability of scientific research institutions based on fusion features.Firstly,author's individual features are constructed using the information of author.Then,A kind of heterogeneous ensemble learning(stacked generalization model)is designed.In the first layer,the fusion of two features and two features are applied to each base model as input.The results of each model are fused as the input of the second layer model,and the final prediction results are obtained by training.Lastly,the single feature and fusion feature are applied to the homomorphic ensemble learning method(random forest model),which is used to predict the influence of the institutions.The predictive effect of institutional influence is well predicted by the integration of institutional centrality and author's individual features in stacked generalization and random forest algorithms,0.7236,0.8495,respectively.For single feature,the accuracy of forecasting by random forest method is also higher than 0.7067,0.7807,respectively.Compared with the other work,this paper improved the prediction performance of institutional influence.