Technology analysis and forecasting using data mining and simulation

다국어 초록 (Multilingual Abstract)

A Technology analysis is increasingly highlighted as socioeconomic growth is strongly dependent on science and technology more and more. A recent trend in the studies on technology analysis is to prescribe how to optimize a target science and technology system to be able to reach the socioeconomic goal by managing the behavior of the system. The conventional studies attempt to describe the behavior of science and technology system and predict its future development in advance.
For the purpose, the studies on technology analysis develop indicators for measuring the status quo of science and technology and describe the behavior of science and technology system that has a certain objective. Based on the indicators, the studies predict the future development of science and technology by discovering a pattern from the dynamic change of the indicators over time and by imitating the behavior of the science and technology system in a mechanistic way. And then, they use the indicators for prioritizing research fields (or, research projects) and selecting some of them with an objective to invest for developing the optimal science and technology policy to meet its socioeconomic goals under the budget constraint. Accordingly, the studies help set up a plan for preparing its future to be able to achieve its socioeconomic goals. It is, however, generally understood that an appropriate prescription of a plan comes from an accurate prediction of the science and technology system, and the prediction follows a description of the system which is appropriate for a policy.
Some empirical studies on science and technology policy (Godin, 2005; OECD, 2007, 2013; Muldur et al., 2007) pointed out some experiences with the failure in prescribing an appropriate policy. They noticed that the current indicators were showing only the past and the present status of science and technology without having the predictive value. Namely, the conventional indicators and methods of technology analysis, which heavily depends on use of the indicators, fail to assess if the behavior of science and technology system achieves its goals.
As a result, the conventional approach to the description and prediction, which aims to prescribe a science and technology policy for achieving its socioeconomic goal, appears to bring forth two sorts of problems as follows: First, the description approach gets into trouble when such description has nothing to do with or lacks its relevancy of the purpose of science and technology policy. Second, the prediction also reveals its shortcoming when it fails to fully account for the behavior of the target science and technology system.
To solve these problems, this dissertation attempts to develop an alternative approach to description and prediction of science and technology. To the end, this study examines the description and prediction approach to science and technology by using the data mining and simulation techniques.
We describe a research product generated by a researcher as a point on a coordinate system of various disciplines. The point represents the position of the research product with respect to multiple disciplines. By locating the research products on a space, we analyze the research behavior of researchers. We define the research behavior by the behavior of a researcher represented by a research product. We also predict the behavior of research products on the coordinate system by developing researcher agent-based modeling and simulation based on the system (i.e., multidisciplinary research space). We call the description and prediction of the research behavior as descriptive research behavior analysis and predictive research behavior analysis, respectively.
Our result shows that the descriptive research behavior analysis on the top of multidisciplinary research space is feasible in detecting a state of art research in convergence. The predictive research behavior analysis based on the descriptive research behavior analysis is good at foresight but an improvement is required for forecasting.
Our contributions are five folds: the evaluation of a research product based on multiple disciplines, an advancement of technology analysis as a mathematically robust science with an orientation to science and technology policy research, the descriptive research analysis by means of text mining with statistical analysis, use of an interactive agent-based intelligent modeling and simulation for a comprehensive and predictive analysis on research behavior, and the advancement in modeling and simulation methodology by the convolution of model-driven or mechanistic approach (agent-based modeling and simulation) and evidence-based approach (data mining).