DRE!: A RULE-BASED SOFTWARE REQUIREMENT EXTRACTION FRAMEWORK FOR INDONESIAN DOCUMENTS

Abstract

Extracting software requirements from descriptive documents is a crucial yet challenging phase in software engineering, par-ticularly when performed manually. This research proposes DRE!, a framework for automatically extracting functional requirements from Indonesian-language system descriptions. The employed methodology applies a concise workflow that begins with data acquisition, involving a system description and a predefined list of actors. The process continues with preprocessing, word dependency parsing for grammatical analysis, actor detection featuring a previous-sentence refer-ence mechanism, and requirement construction using an ac-tor-centric template. The workflow concludes with a similarity check (cosine similarity) for redundancy elimination. The framework was evaluated on eight software description da-tasets. The evaluation results demonstrate promising perfor-mance, with F1-Scores on seven of the eight datasets ranging from 0.76 to 0.88. Peak performance was recorded on the "EduLearn" dataset (F1-Score 0.88). However, an anomaly was identified in the "LogiWare" dataset (F1-Score 0.52), which was attributed to a high rate of False Positives. This finding indicates that DRE! is effective in extracting require-ments, but its performance is highly influenced by the clarity and linguistic style of the source documents.