UltraWiki introduces negative seed entities to enhance ultra-fine-grained ESE, addressing ambiguity and defining "unwanted" semantics. The dataset encompasses 50,973 entities and 394,097 sentences across 236 ultra-fine-grained semantic classes. Two frameworks, RetExpan and GenExpan, are proposed for model evaluation.
The content discusses the challenges of traditional ESE methods in representing ultra-fine-grained semantic classes and introduces negative seed entities as a solution. The UltraWiki dataset is constructed to facilitate research in this area. Two frameworks, RetExpan and GenExpan, are proposed for evaluating large language models on the Ultra-ESE task.
Key points include:
Ke Bahasa Lain
dari konten sumber
arxiv.org
Wawasan Utama Disaring Dari
by Yangning Li,... pada arxiv.org 03-08-2024
https://arxiv.org/pdf/2403.04247.pdfPertanyaan yang Lebih Dalam