NEST: مدل های کمی برای تشخیص روندهای در حال ظهور با استفاده از یک شبکه نظارت متخصص جهانی و شبکه های بیزی

The analysis of changes in the research and development (R&D) environment and developing foresight of future technologies are increasingly recognized as important to support policy decision making and efficient resource distribution. Many futurists are developing foresight of future technologies based on Delphi studies, unfolding history, brainstorming, expert surveys, trend analysis, data mining, and so on. However, formalizing these processes is still a necessary task. In this paper, we introduce the NEST (New and Emerging Signals of Trends) model developed by the Korea Institute of Science and Technology Information (KISTI). The NEST collects information from worldwide expert networks and detects the weak signals of emerging future trends systematically, based on massive data analysis, inference techniques, and Delphi studies, to support the development of foresight of future research and technology. The NEST model combines quantitative and qualitative approaches. In the quantitative approach stages, NEST uses clustering, pattern recognition, and cross-impact analysis using a Bayesian network. In the stages of qualitative approaches, NEST conducts environmental scanning, brainstorming, and a Delphi study.

The importance of utilizing the knowledge created from mass collaboration has become more significant than ever before. Successful collaborative and social networking applications, such as Wikipedia, Twitter, and Facebook, are good examples of this trend. These applications greatly influence our lives including our style of learning, working, communicating, and decision making. Information presented by mass media, articles, news, workshops and conferences is also allowing us to understand our markets, politics, and the global environment [1], [2] and [3]. Predicting the future using this knowledge has long been a research interest of many futurists. Developing foresight of future technology is a promising application of collaborative and social networking, because the essential stages of the foresight process, such as analyzing an R&D environment and inferring future changes, require expert knowledge and a deep understanding of phenomena from all over the world and in all areas, such as politics, sociology, engineering, natural sciences, and religion. Traditional studies of future trends depend mainly on qualitative approaches, such as Delphi studies [4], [5] and [6]. One of the most notable cases of the use of a Delphi study for developing foresight in technology is the case of Japan [7]. Some researchers have used off-line interviews, data analysis [8], data mining [9], literature reviews, and brainstorming [10]. Recently, however, more futurists and information scientists have viewed collaborative knowledge and social networks as a key source for future studies [11], [12] and [13]. The Korea Institute of Science and Technology Information (KISTI) [14], founded in 1962, is a government-funded national information research institute of Korea. KISTI's mission includes archiving scientific information, including academic papers and patents of major countries, scientific factual data, and so on, and operating computing resources such as super-computers and a high-speed R&D network to support scientists. Also, one of its major missions is to analyze scientific information to support the nation's policy makers and decision makers of small and medium sized enterprises (SMEs). A group of researchers in KISTI has focused on forecasting future technology mainly based on Delphi studies and expert’ brainstorming. However, the qualitative approaches of developing foresight, such as Delphi studies, interviews with intellectuals, workshops, and brainstorming, are not sufficient to fulfill the needs of KISTI in several aspects. The first reason is the high cost of conducting the research. Employing intellectuals and experts from all areas of technology and science and conducting online/offline surveys is expensive. The second reason is the slow process speed of research. The interactive procedures of Delphi studies, holding workshops and conferences, and conducting brainstorming take time and it is very difficult to finish the whole process in a year. However, KISTI needs to perform all the processes of developing foresight on an annual bases or anytime a request arrives. The third reason is that the qualitative methods rely on subjective opinions of intellectuals and experts. Based on the experts and intellectuals employed, the results of foresight are different and inconsistent for repeated trials. This aspect is also related with the difficulties that KISTI experienced when answering specific detailed requests of policy makers and SMEs, because even experts and intellectuals have limited knowledge and information to which the can refer. The last reason is that there is a need to utilize environmental monitoring data collected by the Global Trends Briefing (GTB) network [15], which will be described in Section 3.1, and the computing resources KISTI possesses, such as high performance computers, to obtain consistent and fast results. For these reasons, it is necessary to develop a sophisticated systematic model, which combines quantitative and qualitative approaches for developing foresight. Therefore, another group of KISTI researchers is trying to develop a sophisticated systematic foresight development model, the NEST (New and Emerging Signals of Trends) model, which works on the analysis of large volumes of environmental change data to support the first group. In this paper, we introduce KISTI's collaborative environmental analysis model—NEST—using the Bayesian network for developing foresight regarding future technology, which will be useful for supporting the decision-making of groups or nations and for establishing an R&D strategy. This model is designed to find the weak signals of future changes. Weak signals are events, accidents, or strange occurrences that are thought to be the beginning of future changes [16] and [17]. Whilst the concept of weak signals began to be discussed a quarter of a century ago in strategic management literature and its importance has been widely perceived, actual research on modeling the analysis and detection processes still needs to receive significant attention [18]. The objectives of the NEST model include: (1) establishing a systematic process for the identification of weak signals of future trends in research and technology; (2) building a reference system supporting future technology researchers and decision makers; (3) utilizing unstructured data, such as information from mass media, news, conferences, workshops, academic papers, the Internet, and so on, for environmental scanning; (4) utilizing quantitative methods as well as qualitative methods. The purpose of the NEST model is not to automate the process of developing foresight or weak signal detection. Rather the research goal of the NEST model is to build a supplementary information system that provides researchers of future technology and policy-makers with the information about environmental changes and hints of possible weak signals for future changes in research and technology. The concepts of ‘finding weak signals’ and ‘finding future trends’ are used similarly in this paper because the NEST model is a network-based trend detection model in which related signals are connected with each other and weak signals can be found by tracing back the sub-network representing the future trends. The next section presents the results of a literature review on related research. Section 3 presents a detailed description of the NEST model and its components. In Section 4, a Delphi study is performed in order to evaluate the results of NEST, and determination of emerging technology trends is discussed. We then conclude the paper in Section 5.

For about 6 years, the NEST process has been steadily revised, and the results of the top 10 emerging technology trends of each year have been widely announced in Korea by the major public media. Furthermore, the results of NEST research are provided to many governmental organizations to support policy decision making and efficient resource distribution within the country. The NEST model is an effort to formalize the processes of environmental scanning for weak signals and emerging technology trend searches, by combining quantitative and qualitative analysis methods. In this research, the dynamic characteristics of future signals, introduced by Hiltunen's three dimensional concept of the future signal [16], are extended to apply information analysis techniques. The three attributes of future signals—the number of events, the strength of visibility, and the significance perceived by the interpreters—are observed from the GTB data by quantitative analysis using KnowledgeMatrix, a conceptual evaluation index, and a Bayesian network inference model. The weak signals of future trends are determined by tracing back the event network in the signal tracking board used for the Bayesian network inference. Qualitative analysis is also employed in much of the NEST process, such as in refining the environmental data, extracting topics in NEST-clipping, estimating the initial impacts of environmental changes, and determining emerging trends in the Delphi study. This combined approach makes it possible to search the early signals of emerging technology trends without analyzing large volumes of data. The GTB evaluation process, the Knowledge Matrix, the weak signal tracking board, and the conceptual evaluation index are unique systematic processes devised for NEST to support the traditional subjective analysis procedures. Through these stages, NEST can perform emerging technology searches in a more objective and systematic way. Compared with a similar research, the BMBF foresight process in Germany [22], the NEST model has similar processes and steps, including the collection of monitoring data, online survey, clustering for filtering, information analysis technique, and assessment. However, from a different point of view, the BMBF process and NEST are different. One of most distinct differences between the BMBF process and NEST is the source data. While the BMBF process uses interview data collected from an international panel, well-known and acknowledged experts, and workshops, NEST uses data from the GTB collection collected from the mass media, the news, conferences, workshops, and the Internet by an international panel. Members of the NEST panel are not allowed to enter their personal opinions on source data. They are only allowed to perform quality checks regarding to the freshness, duplication, writing quality, and significance of the information found from many sources. The fundamental difference between the NEST and BMBF processes is the manner of driving the whole process. The approach of the BMBF process can be called a supervised top-down approach because the important decisions made during the process, such as the number of topics and start fields, topic selections, and clustering, are made using the knowledge of experts. However, the approach of NEST can be called an unsupervised bottom-up approach because all the intermediate decisions are made by information technology and data mining algorithms. Human experts are involved only in the assessment process at the end of the process. These different features of NEST—preventing experts’ opinions from entering the source data and avoiding an unsupervised bottom-up approach—could be advantageous. In this approach, the possible bias or limits caused by already held human knowledge can be avoided and unexpected important discoveries can be made. However, at the same time, these features cause some limitations. Besides the noise of information, it is hard to control the significance of outputted trends. That is, the output is the subset of the most important future trends on which there is general agreement. Also, in this approach, further analysis and confirmation of qualitative research are essential. Despite these limitations, the NEST system can be useful as a reference system for supporting futurists in developing foresight about future changes in research and technology. Our NEST model research is still in its early stages and the subjective judgments of human experts tend to be the most effective factor of the whole process. Therefore, in order to develop a clearer and more objective weak signal and future trend searching process, it is necessary to perform more in-depth studies, such as formalizing the concept of weak signals and future trends, devising more elaborate and quantitative search methods, defining the index for identifying weak signals and future trends, and developing their applications. In addition, the development of scientometric weak signal and future trend detection processes in the various semi-structured or structureless environmental change data is necessary for future work