Automatic Extraction of User Story Elements: NLP-Based Approach

This study explores the application of Named Entity Recognition to identify user story elements in Indonesian-language texts extracted from software development articles. Two NLP models, tok2vec and transformer, were trained and evaluated on a manually annotated dataset of fifty documents. The transformer model achieved higher validation accuracy, ranging from 75 to 100 percent, while the tok2vec model showed a higher number of detected entities but lower accuracy, with an average of 62.29 percent. This work is limited by several factors, including the small dataset size, lack of domain diversity, absence of baseline comparisons, and the user of a single annotator with inter-annotator agreement. Despite these constrains, the results provide preliminary insights into how NER models can support automated extraction of user story elements in Indonesian software documentation. These findings may assist future efforts to enhance NLP tools for agile requirements analysis in low-resource languages.