From relation extraction to triplet extraction: A survey on low-resource scenarios

Recent advances in generative frameworks have shifted Relation Extraction (RE) toward Triplet Extraction (TE), framing Information Extraction (IE) as an open-schema generation of (subject, predicate, object) triplets. This work provides a structured overview of current solutions, research gaps, and future directions for RE and TE in low-resource scenarios. This survey focuses exclusively on low-resource scenarios, defined by the limited availability of annotated data and/ or the absence of supporting resources such as external vocabularies or ontologies, which are typically leveraged to identify and extract entities and relations. The goal is stated by four key objectives: (1) illustrating the conceptual foundations of RE and TE, including their similarities and differences; (2) formally defining low-resource scenarios and characterizing their key challenges; (3) analyzing State-of-The-Art (SoTA) methods regarding their efficacy in low-resource settings for solving RE and TE; and finally, (4) providing future directions for RE/ TE in such settings.