【CSET】2024紧跟新兴技术趋势利用大数据指导规划研究报告42页

Introduction and Background

This issue brief outlines a human-machine teaming approach for identifying emerging technology trends using big data and AI. Decision-makers face a flood of new research, so this method combines quantitative analysis with subject matter expert (SME) input. Background includes prior research on technology forecasting, such as horizon scanning and Delphi methods, emphasizing the mix of quantitative (e.g., bibliometric analysis) and qualitative approaches for better predictions. The proposed method builds on this by using CSET's merged corpus of scientific publications and the ETO Map of Science, which clusters papers based on citation patterns.

Methodology

The methodology consists of five steps:

1. Identify relevant papers: Start with a set of papers authored by or related to an organization (e.g., DoD-affiliated papers in the proof-of-concept).
2. Locate clusters: Find research clusters in the ETO Map of Science containing these papers, using a dataset from CSET's resources.
3. Analyze metadata: Examine cluster metrics like growth (recent publication rate, normalized for size and age) and export activity (links to other clusters, indicating knowledge transfer).
4. Select and present: Use percentile cutoffs to choose a subset of clusters (e.g., focusing on high-growth or high-export clusters) and share them with SMEs, including metadata like AI classification from a CSET AI classifier.
5. Engage SMEs: Facilitate a discussion based on the clusters, using tools like spreadsheets or web pages from ETO Map, to contextualize data and explore implications (Appendix D provides discussion prompts).

Proof-of-Concept Discussion

In a test with DoD-affiliated papers, the method was applied to identify and analyze clusters with high growth or export activity. Key findings:

• Clusters like those on image forgery detection and robotics were found; SMEs discovered new areas and discussed military applications.
• Benefits included prompting exploration of unfamiliar research and identifying gaps, such as the absence of large language models in the dataset.
• Limitations: The method is speculative due to data lag and may miss cutting-edge innovations (e.g., recent AI breakthroughs), and foundational clusters can be less insightful for SMEs.

Conclusion and Future Applications

The approach supports decision-making by illuminating data trends and enhancing human judgment, but it does not predict the future definitively. Future applications can extend to other fields, such as AI in healthcare, by adapting metadata criteria (e.g., using different classifiers or focusing on application areas). The method highlights advantages like supplementing expert networks, but it has limitations, such as data inconsistencies in funders and language coverage, which could be addressed with more refined tools. Overall, it offers a robust way to manage research overload and foster innovation.