Table 3:

Performance of models under different promptings and data integration methods on traditional NLP metrics. ↑ indicates the higher the better.

	BLEU-1 Avg ↑	BLEU-4 Avg ↑	ROUGE-L Avg ↑	Hallucination (%)↓
qwen2:7b_FS_UGP	0.011	0.000	0.008	92.860
qwen2:7b.ZS-UGP	0.017	0.000	0.011	92.860
qwen2:7b_ZS_UCP	0.280	0.016	0.252	64.740
qwen2:7b_ZS_FCSP	0.281	0.016	0.252	64.680
qwen2:7b_FS_UCP	0.324	0.012	0.299	57.510
qwen2:7b_FS_FCSP	0.326	0.012	0.299	57.320
qwen2.5:7b_ZS_UGP	0.018	0.000	0.012	92.860
qwen2.5:7b_FS_UGP	0.047	0.004	0.038	88.240
qwen2.5:7b_ZS_UCP	0.420	0.025	0.392	44.110
qwen2.5:7b_ZS_FCSP	0.421	0.026	0.391	44.210
qwen2.5:7b_FS_UCP	0.462	0.024	0.427	38.550
qwen2.5:7b_FS_FCSP	0.463	0.025	0.426	38.550
qwen2.5_32b_FS_UGP	0.002	0.000	0.001	92.860
qwen2.5_32b_ZS_UGP	0.018	0.000	0.012	92.860
qwen2.5_32b_ZS_UCP	0.371	0.030	0.343	55.410
qwen2.5_32b_ZSCOT_UCP	0.371	0.031	0.343	55.570
qwen2.5_32b_ZS_FCSP	0.376	0.031	0.346	54.960
qwen2.5_32b_ZSCOT_FCSP	0.376	0.031	0.346	54.960
qwen2.5_32b_FS_UCP	0.383	0.033	0.348	54.330
qwen2.5_32b_FS_FCSP	0.386	0.033	0.350	53.870
llama3:8b_FS_UGP	0.010	0.000	0.007	92.860
llama3:8b_ZS_UGP	0.018	0.000	0.012	92.860
llama3:8b_ZS_UCP	0.338	0.020	0.303	61.160
llama3:8b_ZS_FCSP	0.339	0.021	0.303	61.110
llama3:8b_FS_UCP	0.365	0.024	0.329	53.230
llama3:8b_FS_FCSP	0.367	0.024	0.331	52.970
gpt4o_ZS_UGP	0.003	0.000	0.001	92.860
gpt4o_FS_UGP	0.004	0.000	0.001	92.860
gpt4o_ZS_FCSP	0.091	0.007	0.077	85.160
gpt4o_FS_FCSP	0.092	0.007	0.078	85.160
gpt4o_ZS_UCP	0.093	0.007	0.079	84.880
gpt4o_FS_UCP	0.093	0.007	0.080	84.880