Busro Akramul Umam, Didik Dwi Prasetya, Wahyu Sakti Gunawan Irianto, Mohammad Bhanu Setyawan, Sahlan M. Saleh, Aang Kisnu Darmawan
Assessing the quality of concept map propositions is a significant challenge in education, often manual and subjective. This study aims to address these limitations by providing consistent, automated feedback. ARM has proven effective, but its application to proposition analysis is limited to the literal token level, which ignores the semantic gap. This study first establishes a systematic basis by quantifying the limitations of traditional ARM. We analyzed 850 concept maps from a course, extracting over 10,250 unique propositions. Rules were generated with a Minimum Support (min_sup) of 0.1 and a Minimum Confidence (min_conf) of 0.5. Comparative results show that FP-Growth generates identical rule sets to Apriori but is 44.9% faster (Apriori's average runtime is 12.7 seconds vs. FP-Growth's 7.1 seconds), confirming its efficiency advantage. Rule quality analysis shows that over 30% of rules are redundant variations and Lift is predominantly near 1.0 (scientific_method → research Lift=1.04), validating the failure of literal ARM to capture semantic insights. By rigorously establishing this baseline, our work paves the way and provides important justification for future SBERT-based ARM workflows that can provide meaningful feedback on student understanding. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Universitas Islam Madura, Department of Information Systems, Pamekasan, Indonesia
