語言背後的求救訊號：基於 NLP 的大學生倦怠辨別與預測;Silent Cries in Language: Identifying and Predicting University Student Burnout Using NLP

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98366

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98366

Title:	語言背後的求救訊號：基於 NLP 的大學生倦怠辨別與預測;Silent Cries in Language: Identifying and Predicting University Student Burnout Using NLP
Authors:	鄭欣華;Zheng, Xin-Hua
Contributors:	資訊管理學系
Keywords:	學生倦怠;自然語言處理;情緒分析;主題建模;理工科;Student Burnout;Natural Language Processing (NLP);Sentimental Analysis;Topic Modeling;STEM
Date:	2025-07-24
Issue Date:	2025-10-17 12:41:17 (UTC+8)
Publisher:	國立中央大學
Abstract:	在全球高度重視心理健康與教育永續發展的趨勢下，大學生倦怠(student burnout)逐漸成為高等教育重要議題。然而，現行研究多以量表為主，較少結合自然語言處理(Natural Language Processing, NLP) 技術進行語料分析，亦缺乏針對不同學科群體（如 STEM 與非 STEM）倦怠異質性之探討。為彌補此一研究缺口，本研究以中央大學企業管理系、資訊管理系與資訊工程系學生為對象，依其學科屬性區分為 STEM 與非 STEM 群體，搭配哥本哈根倦怠量表 (Copenhagen Burnout Inventory–Student version , CBI-SS)量化學生倦怠風險。並藉由半結構式訪談文本，建構一套以文本為基礎的學生倦怠辨識機制，探索語言特徵與倦怠風險之關聯，接續以統計與機器學習模型驗證人格特質、學科背景（STEM/非STEM）、性別與語言特徵對倦怠辨識的影響。統計結果顯示，倦怠與五大人格中的親和性在師生關係倦怠上會有差異性，而盡責性會影響學生的個人倦怠以及學業倦怠；除此之外非STEM與STEM學生在師生倦怠上也會有差異性。量化實驗結果在主題建模和情緒分析上，亦可有效了解個體的在特定的倦怠類型。最終在模型辨識上，結合語言特徵與背景變項之模型可達中度辨識準確率，證實語言為辨識倦怠之潛在訊號來源。本研究學術貢獻包括：一、拓展 NLP 應用於心理健康研究之新視角；二、驗證不同學科學生倦怠表現之異質性；三、建構本土化語言資料庫與預測模型，提升語意辨識在輔導與教育情境中的應用潛力。實務上，本研究可作為智慧校園中學生心理預警與輔導介入之參考基礎，亦有助於未來發展多模態數位心理健康監測系統，促進學生自我覺察與資源連結之即時性。;With growing global attention to mental health and sustainable education, student burnout has become a key issue in higher education. However, most research relies on quantitative scales, with limited use of Natural Language Processing (NLP) and little exploration of burnout differences between STEM and non-STEM students.This study addresses these gaps by analyzing students from the Departments of Business Administration, Information Management, and Computer Science at National Central University. Students were grouped as STEM or non-STEM, and their burnout risk was measured using the Copenhagen Burnout Inventory–Student Version (CBI-SS). Semi-structured interview texts were further examined to develop a text-based burnout detection framework. Statistical and machine learning models assessed the impact of personality traits, academic background, gender, and linguistic features on burnout identification.Findings revealed that agreeableness was linked to teacher-student relationship burnout, and conscientiousness was associated with both personal and academic burnout. Differences in teacher-student burnout were also found between STEM and non-STEM students. Topic modeling and sentiment analysis effectively identified burnout types, and a classification model combining linguistic and background features achieved moderate accuracy—supporting language as a potential signal of burnout.This research offers three key contributions: (1) applying NLP to mental health studies, (2) revealing disciplinary differences in student burnout, and (3) constructing a localized corpus and predictive model to support burnout detection. Practically, it lays the groundwork for smart campus mental health monitoring systems and helps promote timely student support and self-awareness.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	34	View/Open

社群 sharing

Loading...