In natural language processing, deep learning algorithms typically rely on large amounts of manually annotated textual data. However, compared to image data, textual data is inherently more ambiguous, often resulting in suboptimal annotation quality. Few-shot learning, which requires only a small number of annotated samples, offers a promising solution to reduce the burden of manual annotation. This paper introduces a few-shot learning approach for entity relationship classification, which achieves global classification by quantifying and matching entity relationship similarities, thereby enabling sample space expansion and migration. The proposed method is evaluated on both Chinese and English entity relationship datasets, with experimental results demonstrating its effectiveness.