Thesis
From raw corpus to lexical-semantic graph: a methodological framework for industrial, agile and high-performance GraphRAG
Teams / Affiliations
Keywords
Abstract (EN)
This project focuses on enhancing access to technical and regulatory knowledge through the integration of Large Language Models (LLMs) and Knowledge Graphs (KGs). In industrial contexts, professionals in domains such as maintenance, legal compliance, and public administration rely on vast and complex documentation. Current information retrieval methods face limitations in scalability, contextual accuracy, and knowledge traceability. The project explores a hybrid approach based on GraphRAG (Retrieval-Augmented Generation with Knowledge Graphs), aiming to combine the generative capabilities of LLMs with the structured, semantically rich representation of domain knowledge provided by KGs. The objective is to automatically extract and organize relevant knowledge from heterogeneous sources (regulations, procedures, tickets, technical documentation) and leverage this structure to support question-answering and decision-making systems. The scientific goals include: (1) dynamic construction and updating of domain-specific knowledge graphs; (2) scalable information retrieval and response generation; (3) semantic reliability and user-personalized outputs; (4) transparent reasoning and multi-source synthesis; and (5) evaluation of the KG's contribution to response quality. This research builds on recent advances in hybrid architectures, and addresses key challenges in graph construction, semantic alignment, and system evaluation. By enabling more contextual, accurate, and explainable interactions with knowledge systems, the project contributes to the modernization of information access in knowledge-intensive industries and advances the field of knowledge-based AI in real-world operational settings.
Contact
You can reach me via LinkedIn.
Curriculum Vitae
You can download my full academic CV here:
Download CV (PDF)