Pre-training to Match for Unified Low-shot Relation Extraction

TL;DR

This paper introduces Multi-Choice Matching Networks and triplet-paraphrase meta-training to unify low-shot relation extraction, effectively handling both zero-shot and few-shot scenarios with significant performance improvements.

Contribution

It presents a novel unified framework and training method that bridge zero-shot and few-shot relation extraction, outperforming existing baselines.

Findings

Outperforms strong baselines on three low-shot RE tasks.

Achieves top results on few-shot RE leaderboard.

Demonstrates effectiveness of triplet-paraphrase meta-training.

Abstract

Low-shot relation extraction~(RE) aims to recognize novel relations with very few or even no samples, which is critical in real scenario application. Few-shot and zero-shot RE are two representative low-shot RE tasks, which seem to be with similar target but require totally different underlying abilities. In this paper, we propose Multi-Choice Matching Networks to unify low-shot relation extraction. To fill in the gap between zero-shot and few-shot RE, we propose the triplet-paraphrase meta-training, which leverages triplet paraphrase to pre-train zero-shot label matching ability and uses meta-learning paradigm to learn few-shot instance summarizing ability. Experimental results on three different low-shot RE tasks show that the proposed method outperforms strong baselines by a large margin, and achieve the best performance on few-shot RE leaderboard.