Structure-Guided Entity Resolution: Fine-Tuning LLMs for Robust Name Matching in Complex Linguistic Contexts
A new framework called Structure-Guided Entity Resolution (SGER) has been developed to improve how Large Language Models (LLMs) match names, particularly in complex linguistic situations. SGER uses a two-phase curriculum to first teach the LLM about name structures and then optimize it for entity matching. This approach achieved 99.02% accuracy and an F1 score of 0.994 on Indian identity data, outperforming existing methods like GPT-4o prompting. The SGER system is now in production at Dream11, a platform serving over 250 million users, demonstrating its scalability and effectiveness in real-world multilingual applications. AI
IMPACT Enhances LLM capabilities for precise name matching in multilingual, real-world systems, crucial for KYC and user identity unification.