NLP-based Topic Modeling and Mining of Heart-to-heart Interview Transcripts: Analysis of Causes for Student Unwillingness to Attend Classes and Implications for Educational Reform
Year: 2026
Type: conference-paper
Abstract: To address the coexistence of overt and hidden absenteeism among vocational college students and the limitation that traditional questionnaires struggle to capture their underlying motivations, this study employs Natural Language Processing (NLP) technology to establish a data analysis pipeline consisting of text collection, de-identification and cleaning, Chinese word segmentation, TF-IDF feature representation, Latent Dirichlet Allocation (LDA) topic modeling, and educational interpretation. This pipeline is used to extract deep-seated topics related to students' unwillingness to attend classes from heart-to-heart interview transcripts. A total of 217 heart-to-heart interview transcripts were collected from a vocational college during the 2025-2026 academic year, forming a text corpus of approximately 100,000 words. After de-identification, spoken language noise removal, Jieba word segmentation, domain-specific dictionary construction, stop word filtering, and vectorization processing, the LDA model was applied for topic mining. The optimal number of topics (K) was determined by comprehensively evaluating three metrics: perplexity (lower is better), topic coherence (higher is better), and human interpretability. Six core topics with high interpretability were identified, including insufficient perceived value of course content, teachers' over-reliance on PowerPoint reading and lack of classroom interaction, and avoidance behavior due to failed self-management. The results indicate that students' unwillingness to attend classes is not merely a disciplinary issue, but a comprehensive outcome of the combined effects of course content, teacher behavior, evaluation mechanisms, students' learning foundations, professional cognition, and environmental support. This study introduces NLP methods into the student management scenario, transforms natural language expressions in heart-to-heart interviews into quantitative clues for classroom teaching reform, and provides empirical support for vocational colleges to carry out precision teaching diagnosis and continuous quality improvement.
Authors: Zhengjie Peng, Wei Wei, Haozhou Yan, ShaoLin Li
Institution: Chongqing Three Gorges Vocational College
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7220867352