From data to design

4,90 €
Auf Lager
Artikelnummer
03654_2024_06-07_03
LLM-enabled information extraction across industries
PDF Download
This paper explores the application of Large Language Models (LLMs) in the automotive and supplier industries, with a particular focus on the use of retrieval-augmented generation (RAG) systems to streamline information retrieval from technical documentation. The research, part of the CoLab4DigiTwin project, investigates how digital twins supported by smart services can enhance interdisciplinary collaboration and reduce the reliance on manual data searches. We developed a pipeline utilizing a RAG architecture which uses a vector database for efficient data management and fast access to relevant information, eliminating the need for expensive computational resources. The performance of various open-source LLMs, which are finetuned on German, was evaluated, focusing on readability, clarity, and accuracy. The results show decent performance of the system without the need for model fine-tuning. Future research will aim to refine these processes and extend the applicability of RAG systems, highlighting the potential of Large Language Models to transform industrial data interaction.
Mehr Informationen
Autoren Robert Becker, Laura Steffny, Thomas Bleistein, Dirk Werth,
Erscheinungsdatum 28.06.2024
Format PDF
Verlag Vulkan-Verlag GmbH
Sprache Deutsch
Seitenzahl 9
Titel From data to design
Untertitel LLM-enabled information extraction across industries
Beschreibung
This paper explores the application of Large Language Models (LLMs) in the automotive and supplier industries, with a particular focus on the use of retrieval-augmented generation (RAG) systems to streamline information retrieval from technical documentation. The research, part of the CoLab4DigiTwin project, investigates how digital twins supported by smart services can enhance interdisciplinary collaboration and reduce the reliance on manual data searches. We developed a pipeline utilizing a RAG architecture which uses a vector database for efficient data management and fast access to relevant information, eliminating the need for expensive computational resources. The performance of various open-source LLMs, which are finetuned on German, was evaluated, focusing on readability, clarity, and accuracy. The results show decent performance of the system without the need for model fine-tuning. Future research will aim to refine these processes and extend the applicability of RAG systems, highlighting the potential of Large Language Models to transform industrial data interaction.
Eigene Bewertung schreiben
Sie bewerten:From data to design
© 2017 Vulkan-Verlag GmbH