Topic-oriented open relation extraction with seed generation

The field of open relation extraction (ORE) has recently observed significant advancement thanks to the growing capability of large language models (LLMs). Nevertheless, challenges persist when ORE is performed on specific topics. Existing methods give sub-optimal results in five dimensions: factualness, topic relevance, informativeness, coverage, and uniformity. To improve topic-oriented ORE, we propose a zero-shot approach called PriORE: Open Relation Extraction with a Priori seed generation. The PriORE leverages the built-in knowledge of LLM to maintain a dynamic seed relation dictionary for the topic (which is initiated a priori). It initially generates seed relations from topic-relevant entity types and can be expanded during extraction. PriORE then converts the more random ORE task to a more robust relation classification task by comparing the relation dictionary to contexts. Experiments demonstrate this approach empowers better topic-oriented control over the generated relations and thus greatly improves ORE performance along the five dimensions, especially on specialized and narrow topics.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Ding, Linyi

Contributors dc:contributor

• Han, Jiawei

• Relation Extraction
• Large Language Models

Statement dc:rights

• Copyright 2024 Linyi Ding

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125722

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125722