Reproduce “Few-Shot, No Problem: Descriptive Continual Relation Extraction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nguyen Xuan Thanh, Anh Duc Le, Quyen Tran, Thanh-Thien Le, Linh Ngo Van, Thien Huu Nguyen
DCRE adds relation descriptions as class anchors alongside replayed examples for few-shot continual relation extraction. During learning it aligns examples with descriptions and at prediction combines description similarity with prototype distance. On BERT and TACRED, its accuracy after task eight is 63.21%, compared with CPL's 57.39%. Accuracy still declines across tasks; the results cover two datasets and two encoder backbones.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.