| Summary: | 博士 === 輔仁大學 === 商學研究所博士班 === 106 === The daily broadening large scale data mounts many new challenges to the processing and analysis of various types of data. Management science is subjected to related data and its inferred consequences. However, the effectiveness of some traditional statistical methods are open to doubt, and the outcomes of induction and deduction may be hard to accurately pilot the related management and operations.
In practical operation, we usually find the relation between variables is more changeable and unreliable than that is assumed in traditional statistical methods; the analysis depends on individual/local data no longer meet people's demands. Analysis based-on network may offer the possibilities for exploring relations between variables and provide systematic perspectives on problems. As the data volume increases, the problem of sparseness will arise. On the measurement of correlation, the traditional methods do not consider the sparseness of the data adequately; simultaneously, whether the correlation is merely a simple linear relationship is still under discussion.
This thesis attempts to improve the problems encountered by the traditional correlation coefficients in the case of non-linear and inflated from the analysis of the correlation between the Copula bases network analysis method. And this method can systematically obtain important factors and categories in the data. The simulation analysis results show that the Copula based correlation measurement can better describe the true relationship between variables with higher robustness and solve the variance and bias under traditional correlation measurement.
Empirical study used the publications of NHIRD which were provided publicly by PubMed. We construct the Copula based network model to find the collaborative network between authors and organizations, and discovered the research topic detection and temporal trend based on Medical Subject Headings (MeSH) terms, so as to provide practical value to related administrative department and attempt to bring the theoretical frame of management science to completion.
|
|---|
