A Unified Taxonomy-Guided Instruction Tuning Framework for Entity Set Expansion and Taxonomy Expansion
CoRR(2024)
摘要
Entity Set Expansion, Taxonomy Expansion, and Seed-Guided Taxonomy
Construction are three representative tasks that can be used to automatically
populate an existing taxonomy with new entities. However, previous approaches
often address these tasks separately with heterogeneous techniques, lacking a
unified perspective. To tackle this issue, in this paper, we identify the
common key skills needed for these tasks from the view of taxonomy structures
– finding 'siblings' and finding 'parents' – and propose a unified
taxonomy-guided instruction tuning framework to jointly solve the three tasks.
To be specific, by leveraging the existing taxonomy as a rich source of entity
relationships, we utilize instruction tuning to fine-tune a large language
model to generate parent and sibling entities. Extensive experiments on
multiple benchmark datasets demonstrate the effectiveness of TaxoInstruct,
which outperforms task-specific baselines across all three tasks.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要