Institutsseminar/2024-09-13

Aus SDQ-Institutsseminar
Termin (Alle Termine)
Datum Freitag, 13. September 2024
Uhrzeit 11:30 – 12:45 Uhr (Dauer: 75 min)
Ort Raum 010 (Gebäude 50.34)
Webkonferenz
Vorheriger Termin Fr 16. August 2024
Nächster Termin Fr 20. September 2024

Termin in Kalender importieren: iCal (Download)

Vorträge

Vortragende(r) Nathan Hagel
Titel LLM4DSLs
Vortragstyp Masterarbeit
Betreuer(in) Sebastian Weber
Vortragssprache Englisch
Vortragsmodus in Präsenz
Kurzfassung The steady rise of generative AI and large language models in software development raises the question of its applicability in model-driven engineering (MDE).

Especially when using textual domain-specific languages (DSL) for modeling, insights into the domain itself and the internals of the DSL are required to create adequate models. To tackle this challenge, we present an updated MDE process aided by LLMs, which removes the necessity of manually creating textual models / writing DSL code manually. Additionally, we implemented tooling to use the updated process in practice. The process is evaluated in a user study and shows that it is faster and more efficient than traditional MDE. Furthermore, although not significantly better rated in perceived usability, 15 out of 18 participants still preferred the new process. Our results indicate that MDE is a good application for LLMs and can benefit from its capabilities.

Vortragende(r) Yavuz Karaca
Titel Validating Completeness in Software Requirements Specifications using Large Language Models
Vortragstyp Bachelorarbeit
Betreuer(in) Lars König
Vortragssprache Englisch
Vortragsmodus in Präsenz
Kurzfassung This thesis explores the use of Large Language Models (LLMs) for validating the completeness of Software Requirements Specifications (SRS). Traditional SRS validation is a manual, time-consuming process that often results in missed requirements, potentially leading to project failures. This study leverages LLMs to automate the identification of incompleteness within SRS documents, focusing on completeness types defined by IEEE standards and Kuchta. The effectiveness of GPT-4 and LLaMA3 in identifying instances of incompleteness was evaluated through a series of experiments using different datasets and various prompting techniques. The F1 score was employed as the primary metric to assess performance.
Neuen Vortrag erstellen

Hinweise