Mathématiques et Informatique Appliquées
du Génome à l'Environnement

 

 

 

YAO Xinzhi

Type
Doctorant.e
Sujet
Knowledge Base-Enhanced Prompting for Relationship Extraction using Large Language Models
Date de début
Date de fin
Encadrant(s)
Claire Nédellec, Robert Bossy
Equipe(s)
Bibliome
Contrat de recherche
DATAIA mobilité internationale étudiante
Directeur.trice (pour les thèses)
Jingbo Xia
Année de soutenance (pour les thèses ou les stages)
2025
Ecole/université (pour les thèses et les stages)
Huazhong Agricultural University, School of Informatics
Description/résumé

The diversity and highly structured nature of the information explicitly provided to the LLM significantly complicate the construction of hard prompts. Building on prior research in event extraction and knowledge injection for LLMs, we will explore various representation strategies. One direction involves comparing formal representations of relationship schemas (including argument types and relationship labels) and explicit Knowledge Base (KB) information, with their verbalized counterparts. Furthermore, the ordering of prompt components, whether by semantic similarity or intrinsic nature, has emerged as a critical factor and will be rigorously assessed.