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

Table 4

F1 scores of different EM models on dirty datasets.

Datasets	DeepER	Magellan	DeepMatcher	MCA	Our model	ΔF1
iTunes-Amazon2	—	46.8	79.4	—	92.6	+13.4
DBLP-ACM2	94.9	91.9	98.1	98.5	99.4	+0.9
DBLP-Scholar2	92.3	92.3	94.7	95.2	95.3	+0.1
Walmart-Amazon2	33.6	37.4	59.2	74.7	84.6	+9.9