Figures & data
Figure 2. GAN-BERT Architecture (Croce, Castellucci, and Basili Citation2020). U, L, F, G, D denote unlabeled data, labeled data, fake labels, generator, and discriminator respectively.
![Figure 2. GAN-BERT Architecture (Croce, Castellucci, and Basili Citation2020). U, L, F, G, D denote unlabeled data, labeled data, fake labels, generator, and discriminator respectively.](/cms/asset/f5f0f67d-e2ad-4b67-9de6-4e3117ab2bfe/uaai_a_2083794_f0002_oc.jpg)
Figure 4. SG-Elect architecture (Riyadh and Shafiq Citation2021), U, L, C denote unlabeled data, labeled data, and combinator respectively.
![Figure 4. SG-Elect architecture (Riyadh and Shafiq Citation2021), U, L, C denote unlabeled data, labeled data, and combinator respectively.](/cms/asset/846957ec-9020-4bd2-a364-f40cf8ecfbee/uaai_a_2083794_f0004_oc.jpg)
Algorithm I: Training process of GAN-BElectra based on SG-Elect (Riyadh and Shafiq Citation2021)
Table 1. Summary of Results (F1-Score, Accuracy, Standard Deviation).
Table 2. Summary of Results (Standard Error, Confidence Interval of Standard Error).
Table 3. GAN-BERT’s accuracy in pseudo label generation across three datasets.
Table 4. Detailed Results for the SST5 Dataset.
Table 5. Detailed Results for the US Airline Dataset.
Table 6. Detailed Results for the SemEval Dataset.
Figure 9. Confusion matrices for SST5 dataset for (a) GAN-BElectra (b) SG-Elect (c) Electra and (d) GAN-BERT.
![Figure 9. Confusion matrices for SST5 dataset for (a) GAN-BElectra (b) SG-Elect (c) Electra and (d) GAN-BERT.](/cms/asset/f643ff37-3571-4d37-a58e-e7a2459df250/uaai_a_2083794_f0009_oc.jpg)
Figure 10. Confusion matrices (with red shades) for the US Airline dataset for (a) GAN-BElectra (b) SG-Elect (c) Electra and (d) GAN-BART and confusion matrices (with green shades) for the SemEval dataset (e) GAN-BElectra (f) SG-Elect (g) Electra and (h) GAN-BART.
![Figure 10. Confusion matrices (with red shades) for the US Airline dataset for (a) GAN-BElectra (b) SG-Elect (c) Electra and (d) GAN-BART and confusion matrices (with green shades) for the SemEval dataset (e) GAN-BElectra (f) SG-Elect (g) Electra and (h) GAN-BART.](/cms/asset/6f6273a5-405a-4829-8d0b-b998bf7a392a/uaai_a_2083794_f0010_oc.jpg)