Table 9.

Balanced Accuracy Performance Summary on Three External Datasets. These datasets come from diverse social media platforms. For each column, the best result is bolded, and the second best is underlined.

	Dataset	Red-Sam	Twt-60Users	SAD
Category	Model	Task #2	Task #2	Task #1
Zero-shot Prompting	${\text{Alpaca}}_{ZS\_\mathit{\text{best}}}$	0.527±0.006	0.569±0.017	0.557±0.041
	${\text{Alpaca-LoRA}}_{ZS\_\mathit{\text{best}}}$	0.577±0.004	0.649±0.021	0.477±0.016
	${\text{FLAN-T5}}_{ZS\_\mathit{\text{best}}}$	0.563±0.029	0.613±0.046	0.767±0.050
	${\text{LLaMA2}}_{ZS\_\mathit{\text{best}}}$	0.574±0.008	0.736 ±0.019	0.704±0.026
	$\text{GPT-3}{\text{.5}}_{ZS\_\mathit{\text{best}}}$	0.506±0.004	0.571±0.000	0.750±0.027
	${\text{GPT-4}}_{ZS\_\mathit{\text{best}}}$	0.511±0.000	0.566±0.017	0.854 ±0.006
Instructional Finetuning	Mental-Alpaca	0.604 ±0.012	0.718±0.011	0.819 ±0.006
	$\Delta -{\text{Alpaca}}_{FT\_vs\_ZS}$	↑ +0.077	↑ +0.149	↑ +0.262
	Mental-FLAN-T5	0.582 ±0.002	0.736 ±0.003	0.779±0.002
	$\Delta -{\text{FLAN-T5}}_{FT\_vs\_ZS}$	↑ +0.019	↑ +0.123	↑ +0.012