Table 2.

Comparison of the performance of the individual model—Distilled Bidirectional Encoder Representations From Transformers—and the combined model based on Dempster-Shafer theory in identifying each sentiment class (positive, neutral, and negative).

Sentiment	Sensitivity (%)			Precision (%)			F1 score			Accuracy (%)
	Individual modela	Combined modelb	Individual model		Combined model	Individual model		Combined model	Individual model		Combined model
Negative	92.8	93.4	77.9		81.7	84.7		87.1	81.3		84.6
Neutral	72.5	77.5	98.7		98.8	83.6		86.8	90.3		91.3
Positive	38.7	54.8	46.2		58.6	42.1		56.6	89		92

^aThe individual model is Distilled Bidirectional Encoder Representations From Transformers [31].

^bThe combined model is the combination of Distilled Bidirectional Encoder Representations From Transformers (BERT) [31], Robustly Optimized BERT Pretraining Approach (RoBERTa) [35], BERTweet [38], and fine-tuned RoBERTa [35].