Edouard was a Master Researcher working on Natural Language Processing and Large Language Models (LLMs). His research focused on evaluating knowledge within LLMs and extracting facts from their parametric memory.
Edouard developed ReWiSe, a novel approach that combines chain-of-thought reasoning with relation-wise self-consistency to construct knowledge graphs using LLMs without external corpora. His work on this challenge demonstrates how strategic prompting and self-consistency techniques can significantly improve knowledge extraction from language models. ReWiSe won the 2025 LM-KBC competition with a Macro-F1 score of 44% (compared to a 21% baseline).
His research addressed the critical challenges of LLM hallucination and stochastic generation through innovative reasoning strategies tailored to relation cardinality and schema. The implementation of ReWiSe is available on GitHub.
Search for Edouard Albert-Roulhac's papers on the Research page