Machine learning is located at the center of the 4th industrial revolution, and is changing not only artificial intelligence that grasps information from photos and understands human language, but also the paradigm of existing industries such as manufacturing, health care, and management.
Therefore, companies are actively using machine learning technology to provide a better user experience for customers who use their products. Leading domestic and foreign IT companies such as Google and SKT are trying to provide better voice recognition services for AI speakers, and Samsung is challenging the zero defect rate of its products by using image processing technology and sensors in its semiconductor production line.
In this situation, the best strategy for a company to gain an edge over its competitors is to preoccupy technology through securing new machine learning technology. Accordingly, companies are trying to predict the promising technologies of machine learning.
Existing promising technology predictions include a qualitative method that relies on expert opinions and a quantitative method based on analysis techniques on data such as patents. Since the qualitative method reflects the subjectivity of researchers, recently, quantitative analysis techniques based on patent data have been widely used.
However, patent data takes at least a year from research to registration. Accordingly, the patent-based technology prediction method has a problem in that it does not reflect the latest research conducted in recent months. So, in this study, we focused on open source, new data that reflects research progress immediately.
- Research Purpose: Predicting promising technologies in the machine learning industry using open source data
- Research data: GitHub repository
- Model: Variational Graph Auto-Encoder among graph neural networks