本研究旨在探討深度學習在各國的發展時間與成長趨勢,以及在相關學科 與應用領域上之發展狀況。研究採用專利計量分析法與內容探勘工具 CATAR, 針對美國專利及商標局 1976 至 2018 年之深度學習領域專利進行分析。研究結 果分為四個面向:(1)專利成長趨勢與技術生命週期;(2)專利數分析與趨 勢分析;(3)專利引用分析;(4)專利主題與專利關聯度分析。 研究發現:(1)深度學習領域的技術生命週期正處於成長階段,其專利 申請與公告的延遲時間平均約為 1.75 年;(2)高生產力專利權人國別依序為 美國、日本、以色列、韓國、中國、德國以及加拿大,這七個國家的專利就佔 了整體的 93%,專利高生產力發明人國籍依序為美國、中國、韓國、以色列、 日本、印度以及加拿大;(3)在 103 組專利權人與專利發明人國家組合中, 有 78 組與美國有關;(4)主要引用的學科領域為深度學習、神經網絡以及語 音識別;(5)應用領域有語音識別、影像分析、圖像識別、醫學圖像、以及 車輛控制系統等;(6)臺灣可以參考與學習以色列與韓國的發展模式,在研 究領域方面,臺灣可以往醫學圖像與診斷、外科以及鑑定這個方面多加琢磨。 研究建議:(1)增加關鍵詞(2)針對不同面向進行更深入與更具主題性的研 究(3)針對深度學習領域之研究論文進行研究。
The purpose of this study was to explore the development and the growth trend of deep learning in different countries. Also, the situation of deep learning in other related subjects and the application in different fields. This study used patent analysis and the content mining tool - CATAR to analyze the patents in the field of deep learning from 1976 to 2018 searching from USPTO. The findings of this paper are as follows: (1) The technology life cycle of deep learning is in the growth stage, and on average, the issue date is 1.75 years later than the applied date. (2) On patent assignee's nationality, the countries of high productivity are the US, Japan, Israel, South Korea, China, Germany, and Canada, and patents in these countries account for 93% of the total. On patent inventor's nationality, the countries of high productivity are the US, China, South Korea, Israel, Japan, India, and Canada. (3) Among 103 national groups of assignees and inventors, there are 78 groups related to the US. (4) Citations are mainly related to deep learning, neural networks, and speech recognition. (5) Applications focus on speech recognition, image analysis, image recognition, medical image, and vehicle control systems. (6) Taiwan can learn from Israel and South Korea, and research on medical image. Based on the findings of this study, there are three suggestions: (1) Add keywords. (2) Research on specific subjects intensively. (3) Research on papers in the field of deep learning.