Scientometric Studies Based On Feature Space

Scientometrics is a significant subject category, which quantitative analyzes the process of scientific communication. It focuses on the flow of knowledge and information in the process of communication, and analyzes a series of issues for it. In the traditional research, the mult i- dimensional information of the research object is compressed into a single dimension for measurement, and it cannot meet the demand of more and more elaborate and targeted research. This paper presents a research method based on feature space. It deep mines the characteristics and regularity of scientific activities and information exchange process, after building the feature space for describing the research issues. Through researching several typical issues related to scientometrics for the different objects, the new method is proved to be effective and universally applicable. And the more abundant research category of scientometrics is shown.Firstly, a research method based on fea ture space is presented. In this method, the multi-dimensional information contained by research objects is classified, and the features related to research issues are extracted and kept. And a general methodology is established after introduced the description method based on feature space.Secondly, the citation behavior is analysed in depth and the influence of scientific papers is evaluated. Respectively, the width feature space, depth feature space and importance feature space are constructed. Then the features are weighted by using Analytic Hierarchy Process. And the citation behavior and the influence of scientific papers is analysis.Thirdly, the citation impact based on feature space for describing the influence of scientific papers is predicted. After established a paper‘s feature space, correlation analysis is used to obtain the relationship between citation impact and the features of papers. Then the stepwise multiple regression analysis is used to select appropriate features from the space and to build a regression model for explaining the relationship between citation impact and the chosen features. Several significant features are also analysed in depth.Fourthly, an effective method is developed for the accurate identification of journals that engage in coercive self-citation. The feature space for describing journal citation behavior is established and feature selection is conducted. Then the journal classification model is constructed using the logistic regression method to identify normal and abnormal journals. The performance of the classification model is also evaluated.Fifthly, the process of interaction between subject categories is quantitatively detected. By using citation matrix among subjects a quantitative methodology is presented to investigate the development of subject categories. Respectively, the Dimensions, strength and speed of the knowledge flow are selected to establish the feature space for describing the development. Then the development process of different type of subject categories is analysed.