Real-world relation distributions are long-tail: Only the common relations obtain sufficient training instances and most relations have very limited relational facts and corresponding sentences. Many relations even have less than 10 training instances. The general idea of few-shot models is to train good representations of instances or learn ways of fast adaptation from existing large-scale data, and then transfer to new tasks. 

Performing More Efficient Learning: Methods

Lots of long-tail relations only contain a handful of training examples. However, it is hard for conventional RE methods to well generalize relation patterns from limited examples like humans. Therefore, developing efficient learning schemas
to make better use of limited or few-shot examples is a potential research direction.

Hamilton College

Feasible Approaches Leading to Better RE
Abilities

Han, X. (n.d.). More Data, More Relations, More Context and More Openness:
A Review and Outlook for Relation Extraction *, Retrieved June 4, 2022, from https://arxiv.org/pdf/2004.03186.pdf

More Relations, More Context and More Openness:
A Review and Outlook for Relation Extraction

Supervised RE methods heavily rely on expensive human
annotations, while distant supervision introduces more auto-labeled data to alleviate this issue. Designing schemas to obtain high-quality and high-coverage data to train robust RE models still remains a problem to be explored.

Utilizing More Data

Performing More Efficient Learning

Many relational facts are expressed in complicated context (e.g. multiple sentences or even documents), while most existing RE models focus on extracting intra-sentence relations. To cover those complex facts, it is valuable to investigate RE in more complicated context.

Handling More Complicated Context

Conventional RE frameworks are generally designed for pre-defined relations. Therefore, how to automatically detect undefined relations in open domains remains an open problem.

Learn Before

Related

Learn After