This is the authors' abstract. We don't add key points for this paper.
Few-shot relation extraction is a subtask of relation extraction, aiming to identify new relations by learning from a small number of labeled instances. Most existing few-shot relation extraction methods mainly rely on fine-tuning techniques of pre-trained language models combined with traditional data augmentation methods, which fail to effectively address the issue of insufficient instances, leading to difficulties in improving the performance of few-shot relation classification tasks. Inspired by the latest advances in large language models (LLMs) such as ChatGPT, we propose a few-shot relation extraction method based on data augmentation with large language models and contrastive learning (CLDA-FSRE). This method utilizes large pre-trained models for data augmentation and employs prompt-based learning to guide the large language models, enabling the model to exhibit strong domain adaptation capabilities even in scenarios with limited data. To further mitigate the overfitting problem under few-shot conditions, we introduce a contrastive learning module and a dynamic task weighting mechanism to focus on the learning of difficult tasks. Our experimental results are generally consistent with our design expectations, demonstrating the effectiveness of this method on the FewRel 1.0 and FewRel 2.0 datasets.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: