<![CDATA[The existence of practical evaluation instruments is a vital pillar in the digital learning ecosystem. Automatic Question Generation (AQG) technology is designed to reduce educators' cognitive load, but low keyword extraction accuracy often degrades the quality of generated questions. Single-frequency approaches are susceptible to noise absorption, while graph algorithms ignore word scarcity. This study formulates a hybrid optimization technique by integrating Term Frequency-Inverse Document Frequency (TF-IDF) scores and TextRank centrality to select keywords in Internet of Things (IoT) literature. The processing is executed using an unsupervised learning framework at the unigram level, evaluated against an expert Ground Truth and the grammatical feasibility of cloze tests. Computational calculations at the top 30 word limit proved that the hybrid method achieved an F1-Score of 40.68\%, equivalent to pure TF-IDF and surpassing TextRank (37.29\%). In the pedagogical validation phase, the proposed model demonstrated absolute superiority with a feasibility rate of 86.67\% (26 valid quizzes), subjugating TF-IDF which stalled at 83.33\%. The intervention of graph topology successfully deterred the intrusion of meaningless words and maintained syntactic integrity, making this hybrid formulation highly recommended as the driving engine for future question banks.]]>
Copyrights © 2026
