Leveraging Pretrained Language Models for Enhanced Entity Matching: A Comprehensive Study of Fine-Tuning and Prompt Learning Paradigms

Table 6

F1 scores of different paradigms on dirty datasets.

Datasets	FT	HP/ΔF1	SP/ΔF1
iTunes-Amazon2	90.9	91.5/+0.6	92.6/+1.1
DBLP-ACM2	98.5	99.1/+0.6	99.4/+0.3
DBLP-Scholar2	94.9	95.0/+0.1	95.3/+0.3
Walmart-Amazon2	81.8	83.2/+1.4	84.6/+1.4