Jia-Kai
National Taitung University Department of Computer Science and Information Engineering Student
E-mail: shijiakai77@gmail.com
Yi-Ting You
National Taitung University Department of Special Education Student
E-mail: 11300701@gm.nttu.edu.tw
Abstract
This study aimed to analyze enrollment trends and thematic preferences in digital learning courses using the open dataset from Taipei e-Campus (Taipei e-University) for the year 2018. The study compared two classification methods—the original administrative classification and keyword-based semantic classification—to determine which method more accurately reflects learners’ actual preferences and needs. Using decision tree analysis, results indicated that keyword-based classification (Accuracy =0.83, F1 = 0.80) significantly outperformed the original administrative classification (Accuracy = 0.52, F1 = 0.44), effectively identifying popular course themes such as environmental education and technology applications. Conversely, the original classification system, constrained by administrative structures, failed to accurately predict several categories such as management and language courses. Additionally, certain potential themes, like health and wellness and indigenous culture, faced classification challenges due to limited course offerings and inconsistent naming conventions. The findings suggest future course planning should adopt more detailed keyword-based classification strategies to enhance effectiveness and better respond to adult learners’ educational demands.
Keywords：Lifelong learning、Supervised learning、Adult educational needs、Keyword-based classification