Applied Domain Adaptation of LLM-based Document Embeddings for Engineering Knowledge Retrieval
Abstract
Technical documentation related to engineering and manufacturing is often stored in unstructured documents, rendering search and information retrieval processes time-consuming and inefficient. Recently, Retrieval-Augmented Generation (RAG) pipelines and AI agents leveraging dense retrieval capabilities of document embedding language models have emerged as promising solutions to address these challenges. However, these methods typically lack semantic understanding of industry-specific terminology, limiting their retrieval accuracy in specialized domains. In this paper, we identify the fine-tuning of dense retrieval models within RAG systems as an effective and cost-efficient approach for enhancing their industry- and enterprise-specific knowledge. We further investigate whether pretraining based on Masked Language Modeling is necessary for these scenarios. Through empirical evaluation, we show that automatic data preprocessing and synthetic data generation derived from human-labeled retrieval examples provide a cost-effective strategy for preparing fine-tuning data. By training and evaluating an embedding model on 21 281 engineering documents encompassing best practices and lessons learned, totaling 75 million whitespace delimited words, we achieve a significant increase of \(23\%\) in Recall@10 for domain-specific retrieval performance, while retaining generalization capabilities on general retrieval tasks as measured by the Massive Text Embedding Benchmark (MTEB).…
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.