Few-shot relational triple extraction with hierarchical prototype optimization

Chen Gao, Xuan Zhang, Zhi Jin, Weiyi Shang, Yubing Ma, Linyu Li, Zishuo Ding, Yuqin Liang

Abstract: Highlights•Mitigate prototype bias based on a small number of samples by using prompt learning.•Construct span-based entity-level prototypes to avoid label dependency.•Introduce hierarchical contrastive learning to jointly refine the metric space.