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

Table 3

F1 scores of different EM models on structured datasets.

Datasets	DeepER	Magellan	DeepMatcher	MCA	Our model	ΔF1
BeerAdvo-RateBeer	72.7	78.8	72.7	80.0	86.7	+6.7
iTunes-Amazon1	—	91.2	88.0	—	92.9	+1.7
DBLP-ACM1	97.6	98.4	98.4	98.9	99.2	+0.3
DBLP-Scholar1	92.3	92.3	94.7	95.2	95.7	+0.4
Amazon-Google	62.1	49.1	69.3	71.4	76.2	+4.8
Walmart-Amazon1	39.0	71.9	66.9	74.7	84.1	+9.4
Abt-Buy	36.1	43.6	62.8	69.3	80.0	+10.7