eISSN 2093-744X
pISSN 1598-2645

Table 1

Performance of each individual operation in EDA

Method	1%	10%	100%	Average
EDA	67.1	87.1	91.3	81.8
Deletion	67.5	86.7	91.0	81.7
Swap	67.4	86.1	90.8	81.4
Replacement	67.2	86.3	90.9	81.4
Insertion	67.4	87.0	91.3	81.9

The bold font indicates the best performance in each test. We used the SST-2 dataset to generate four samples for each sentence. We fine-tuned BERT with the augmented dataset for classification. We performed the experiments with various fractions of the datasets, 1%, 10%, and 100%. In the case of 1%, we use only 1% of the dataset for augmentation and model training.

International Journal of Fuzzy Logic and Intelligent Systems 2024;24:83~92 https://doi.org/10.5391/IJFIS.2024.24.2.83