Konferenzbeitrag
Experiences with Using a Pre-Trained Programming Language Model for Reverse Engineering Sequence Diagrams
Volltext URI
Dokumententyp
Text/Conference Paper
Dateien
Zusatzinformation
Datum
2023
Autor:innen
Zeitschriftentitel
ISSN der Zeitschrift
Bandtitel
Verlag
Gesellschaft für Informatik e.V.
Zusammenfassung
Reverse engineering software models from program source code has been extensively studied for decades. Still, most model-driven reverse engineering approaches cover only single programming languages and cannot be transferred to others easily. Large pre-trained AI transformer models which were trained on several programming languages promise to translate source code from one language into another (e.g., Java to Python). Thus, we fine-tuned such a pre-trained model (CodeT5) to extract sequence diagrams from Java code and examined whether it can perform the same task for Python without additional training.