Development and Evaluation of an NLP-Based Key Term Highlighting Model in Digital Learning Environments

다국어 초록 (Multilingual Abstract)

This study proposes and evaluates a key-term highlighting model designed to support learners in digital learning environments. The model automatically extracts and classifies domain-specific key terms from educational content, visually emphasizing them to guide learner attention and enhance comprehension. Text preprocessing, morphological analysis, and candidate-term extraction were performed using KoNLPy with the Okt and Kkma analyzers, combined with TF-IDF-based N-gram vectorization. Candidate terms were validated by verifying the existence of corresponding Wikipedia entries and by measuring Jaccard similarity against an information-subject glossary, which enabled classification into general and domain-specific terms. Highlighted terms were visualized in HTML format to enable learner interaction. Expert validation involved seven specialists in information education who labeled the extracted terms, thereby providing ground-truth data for model evaluation. Phase 1 yielded a precision of 0.84, recall of 0.86, F1 score of 0.85, and accuracy of 0.90. After incorporating expert feedback and refining labels, Phase 2’s performance improved substantially, achieving a precision of 0.91, recall of 0.97, F1 score of 0.94, and accuracy of 0.95. The results demonstrate that the proposed model effectively distinguishes one- and two-word key terms, supports learner attention and comprehension, and alleviates cognitive overload in visually complex digital materials. Differences among expert judgments highlight the need for standardized definitions of domain-specific terminology to ensure consistent learning outcomes. This study demonstrates that combining search-engine verification with glossary-based evaluation enhances AI-assisted educational tools, offering both practical utility in digital learning environments and a foundation for developing high-quality datasets for future AI-assisted applications