미래 기술을 예측하는 것은 정부가 예산을 편성할 때나 기업이 경영 전략을 세울 때 꼭 필요한 정보 중 하나다. 논문 데이터베이스는 일반적으로 미래 기술을 예측하는 데 사용되며 대표적인 논문 데이터베이스는 Web of Science와 SCOPUS입니다. 빅데이터와 인공지능 기술이 발전함에 따라 미래 기술 예측에도 데이터 기반 머신러닝 기법이 적용되고 있다.
한국과학기술정보연구원(KISTI)은 빅데이터와 인공지능을 기반으로 한 미래기술 예측기법을 다년간 연구해왔다. KISTI가 지금까지 수행한 연구는 논문 인용 비율을 주로 활용해 클러스터를 찾아내고 이를 신기술로 해석한다. 최근 유럽연합 집행위원회의 공동 연구 센터(JRC)는 데이터 분석을 기반으로 한 진보된 형태의 과학 기술 약한 신호 검색을 제시했습니다.
이는 과학기술 분야의 빅데이터 분석을 기반으로 하며, 약한 신호를 찾는 진보된 방법을 보여준다. 우리는 전문가의 도움 없이 약한 신호를 자동으로 감지하는 기술을 개발하기 위해 노력했습니다.
Introduction
He used the term "weak signal" as a small signal from outside that can cause a sudden impact13). In the past, the search for weak signals was when a group of experts determined whether there was a weak signal based on news. However, quantitative data analysis is still used as basic information for expert groups to select weak signal candidates.
Furthermore, since the goal of the analysis is often limited to a specific predefined field, it is difficult to see the result as a weak signal in the true sense.15). Weak signal EC is a method of using data analysis - which was only used as a baseline - as a central role in the weak signal detection process. However, even in the EC weak signal detection process, the final weak signals were selected to reflect the process of expert insights such as manual filtering, custom indicators and keyword clustering in the second half of the process.
Therefore, it is believed that a large group of experts should review the weak signals candidate list and select the final weak signals. Although data analysis plays an important role in the method of detecting weak signals, there is still a dependence on experts.
Weak Signal Automated Detection Process
The results selected in step 2 were keywords that showed rapid growth recently, and we called them pop keywords. 2> Keywords that recently appear or show increasing activity in the reference keyword set of paper text corpus are called popping keywords. As a result of analyzing the characteristics of the popping keywords, we have found a difference between the emerging popping keywords (activeness=1) and the remaining popping keywords (activeness<1), i.e. popping keywords with a lifetime of three years or less and those with a lifetime of more than three years.
It is interpreted that the robustness of the newly emerged popping keywords as research subjects is weaker than the existing popping keywords. The group of the popping words included in connected components is defined as a weak signal for the newly emerged popping keywords with a lifetime of less than or equal to three years, extracted clicks that are completely connected on the network with related popping keywords with a spacing of 0.3. A network of related keywords within a distance of 0.3 is a large network connected as a whole.
A group of outlier keywords included in the final click after the clustering process is defined as a weak signal. A total of 439 weak signals were generated based on regular and emerging keyword popping.
For common keywords that appear longer than three years and related keywords that appear within a distance of 0.1, we extracted fully related components separated from other appearing keywords and correlated them like a network. If fully related cliques are extracted from this network, some cliques may redundantly include duplicate keywords. Clicks with duplicate keywords are merged into a group and the merging is repeated until there are no more duplicate keywords appearing between the groups.
Starting with the keywords extracted in phase 1 to generating weak signal candidates in phase 3, the characteristics. some weak signals were incomplete, the algorithm produced all results. The parts that need to be corrected and completed will be reflected in the next upgrade of the automated weak signal detection algorithm. A group that is completely separate from other popping keywords and whose popping . keywords are connected as one network Excluded when the size of connected components is greater than or equal to 100.
If there are duplicate pop keywords between cliques, the cliques are merged. Repeat until no more duplicates exist.
Weak Signal Dynamics
Introduction
The weak signal results of the two years have been accumulated, so that they can be analyzed comparatively. By comparing the bang keywords that make up weak signals, we can find the weak signals that remain the same after one year and understand the changing patterns of weak signals, such as disappeared weak signals, newly emerged weak signals, and changed weak signals. Weak signals, which remain the same this year as last year, can be seen as having a lifetime of two years or longer than weak signals.
If there were changes in the popping keywords that constituted weak signals from last year and this year, it would be helpful to understand detailed changes in technologies by examining which new popping keywords were entered and which popping keywords was disappeared. Combining several weak signals would help us understand the relationship between technologies. As the weak signal results detected each year are accumulated, the lifetime and mechanism of weak signals can be analyzed.
In this report, we first analyze the underlying dynamics using weak signals in 2022 and weak signals in 2023.
Comparative Analysis of Weak Signals 2022 and 2023
Those that were not detected last year, but were detected for the first time this year, are called weak newborn signals. The most weak newborn signals were detected in the fields of medicine (54) and computer science (36). Among the weak signals in medicine, [long-covid, long-covid-19] is one of the weak signals of a newborn.
Among the weak signals in Computer Science, [space-air-ground integrated network, space-ground integrated network] is one of the newborn weak signals. We have found that some popping keywords in some weak signals from last year were replaced with new popping keywords this year. For 102 weak signals, new popping keywords joined the weak signals from last year, and some popping keywords disappeared.
In some of the weak signals detected this year, we found that two or three weak signals from last year, or part of them, were merged and formed a new weak signal. Three weak signals from [p2p lending, p2p lending platform], [equity crowdfunding, crowdfunding platforms, civic crowdfunding, charitable group funding, equity crowdfunding, crowdfunders], and [Islamic banking finance, Islamic financial literacy] in Computer Science have been merged into one weak signal, which indicating that the field is expanding.
Result