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

Table 5

F1 scores of different paradigms on structured datasets.

Datasets	FT	HP/ΔF1	SP/ΔF1
BeerAdvo-RateBeer	80.0	83.9/+3.9	86.7/+2.8
iTunes-Amazon1	91.5	91.5/0.0	92.9/+1.4
DBLP-ACM1	98.6	98.8/+0.2	99.2/+0.4
DBLP-Scholar1	95.7	95.6/−0.1	95.7/0.0
Amazon-Google	73.8	74.1/+0.3	76.2/+2.1
Walmart-Amazon1	79.9	83.5/+3.6	84.1/+0.6
Abt-Buy	74.1	79.0/+4.9	80.0/+1.0