深度學習預測大學教授升等的職級:書目計量學

Deep learning to predict the academic ranks in professor classifications using featured variables of research achievements: a bibliometric analysis

許多大學每年接受很多教師提出教授升等的申請。如何建立客觀篩選機制，成為一項必要且重要的任 務。於某大學網站下載 63.266 期刊論文摘要及其2.128 學者的人口學變項，如學術等第、年資、被引用 論文數、及論文篇數等。利用書目計量學的h指數評量個人研究成就，金字塔圖及拔靴法比較教授等級 間h指數的差異。利用探索性因素分析及羅輯斯迴歸分析挑選適當的特徵變數。再用卷積神經網路（CNN） 模式預測教師是否符合教授的學術資格。結果顯示，教授等級的h指數呈現統計顯著差異。模式預測準 確度達0.70，ROC曲線下面積為0.75。CNN 的程式模組可用來預測教授的等級。我們提供一具評估大學 教師學術成就的方類預測，所提供的CNN 程式模組，提供其他大學未來篩選教授升等之參考。

Many universities receive numerous applicants annually for academic titles(e.g., professor). How to set an objective criterion for screening the eligible candidates is required for the development used in academics. We downloaded 63,266 journal articles along with the corresponding citations from one of the Taiwan national university website on December 10, 2019. A total of 2,128 researchers' biographical characteristics including academic ranks(AR), work tenure, citations, number of published articles were collected. The individual research achievements(RA) were assessed by the bibliometric h-index. The pyramid plot and the bootstrapping method were applied to compare the difference in titles. The exploratory factor analysis and the multiple regression analysis were performed to select the featured variables used in the model of convolutional neural networks(CNN) for predicting ARs. We found that the difference in h-index certainly exists in ARs. The accurate prediction rate reaches 0.70 in discriminating ARs between types of professors and non-professors. The arear under the receiver operational curve(AUC) is 0.75. An APP using the CNN model was created and demonstrated in this study. We provided a CNN prediction tool for screening the applicants who want to understand whether they are eligible for promotion in ARs as a professor(or not a professor) in academics. The CNN APP is worth application for use in other universities in the future.