ECIR-2024-llm-screening

This file serves as a appendix table for result in Section 6: Discussion and outlook; only pure results are showed without further explaination.

Table 1. Uncalibrated results comparison when inclusion-exclusion criteria is used in the prompt (note only four datasets are available here: CLEF-2017, CLEF-2018, CLEF-2019-dta, CLEF-2019-Intervention), Seed Collection does not contain extractable information. Inclusion-exclusion criteria extracted using protocol file, topics that does not contain protocol files are disregarded. 2-tailed t-test with bofferoni correction (p<0.05) are compared with LlaMa-2-7b-chat.

	Model	P	R	B-AC	F3	Suc	WSS
CLEF-2017	BioBERT	0.06	0.95	0.61*	0.30	0.74*	0.26*
	LlaMa-7b	0.04	0.92	0.48*	0.24	0.46	0.03*
	LlaMa-2-7b	0.07	0.50*	0.60*	0.23*	0.02*	0.70*
	LlaMa-2-13b	0.04	1.00*	0.50*	0.25	0.98*	0.00*
	Falcon-7b	0.05	0.92	0.52*	0.25	0.44	0.12*
	Alpaca-7b	0.04	0.92	0.51*	0.25	0.38	0.11*
	Guanaco-7b	0.04	1.00*	0.50*	0.25	1.00*	0.00*
	LlaMa-2-7b-chat	0.08	0.87	0.72*	0.35	0.26*	0.56*
	LlaMa-2-13b-chat	0.19*	0.41*	0.66	0.31	0.04*	0.91*
CLEF-2018	BioBERT	0.06	0.97*	0.59*	0.29	0.87*	0.19*
	LlaMa-7b	0.05*	0.92	0.48*	0.25	0.33	0.04*
	LlaMa-2-7b	0.07	0.49*	0.59*	0.22*	0.03*	0.69*
	LlaMa-2-13b	0.05*	1.00*	0.50*	0.26	1.00*	0.00*
	Falcon-7b	0.05*	0.92	0.51*	0.25	0.40	0.11*
	Alpaca-7b	0.05*	0.91	0.51*	0.25	0.30*	0.11*
	Guanaco-7b	0.05*	1.00*	0.50*	0.26	1.00*	0.00*
	LlaMa-2-7b-chat	0.09	0.88	0.75	0.37	0.27*	0.59
	LlaMa-2-13b-chat	0.26*	0.36*	0.66*	0.30	0.00*	0.94*
CLEF-2019-dta	BioBERT	0.07	0.99*	0.58	0.30	0.88*	0.18
	LlaMa-7b	0.07	0.93	0.48*	0.27	0.25	0.03*
	LlaMa-2-7b	0.08	0.48*	0.58	0.23	0.00	0.68*
	LlaMa-2-13b	0.07	1.00*	0.50*	0.28	1.00*	0.00*
	Falcon-7b	0.07	0.95	0.54*	0.29	0.50	0.12*
	Alpaca-7b	0.07	0.91	0.52*	0.28	0.25	0.12*
	Guanaco-7b	0.07	1.00*	0.50*	0.28	1.00*	0.00*
	LlaMa-2-7b-chat	0.09	0.92	0.71	0.35	0.62	0.49
	LlaMa-2-13b-chat	0.19	0.49*	0.69	0.32	0.00	0.87*
CLEF-2019-Int	BioBERT	0.10	0.98*	0.58*	0.32	0.90*	0.16*
	LlaMa-7b	0.05*	0.86	0.47*	0.26	0.30	0.08*
	LlaMa-2-7b	0.08	0.30*	0.55*	0.18*	0.05*	0.80*
	LlaMa-2-13b	0.05	1.00*	0.50*	0.29	0.97*	0.00*
	Falcon-7b	0.05*	0.91	0.50*	0.27	0.57	0.09*
	Alpaca-7b	0.05*	0.87	0.49*	0.27	0.30	0.12*
	Guanaco-7b	0.05	1.00*	0.50*	0.29	1.00*	0.00*
	LlaMa-2-7b-chat	0.08	0.90	0.70	0.35	0.42	0.48
	LlaMa-2-13b-chat	0.17*	0.45*	0.67	0.33	0.05*	0.87*

Table 2. Calibrated results comparison when inclusion-exclusion criteria is used in the prompt (fused method unavilable due to inclusion-excllusion criteria contain too much token for BioBERT model),2-tailed t-test with bofferoni correction (p<0.05) is compred with Uncalibrated setting.

Dataset	Model	Setting	P	R	B-AC	F3	Suc	WSS
CLEF-2017	7b-chat	Unc	0.049	0.728	0.533	0.245	0.400	0.329
		Cal (0.95)	0.060*	0.921	0.671	0.300	0.500	0.410*
		Cal (1)	0.050*	0.981	0.613	0.270*	0.925	0.239
	13b-chat	Unc	0.193	0.270	0.518	0.228	0.000	0.755
		Cal (0.95)	0.067*	0.933*	0.693	0.311*	0.500*	0.438
		Cal (1)	0.054*	0.978	0.613	0.273*	0.825*	0.240
CLEF-2018	7b-chat	Unc	0.102	0.880	0.711	0.356	0.567	0.521
		Cal (0.95)	0.106*	0.885*	0.731*	0.368	0.500	0.555
		Cal (1)	0.065*	0.975*	0.623*	0.311	0.900	0.258
	13b-chat	Unc	0.333	0.362	0.666	0.321	0.000	0.958
		Cal (0.95)	0.106	0.918*	0.748	0.381	0.500*	0.557*
		Cal (1)	0.069	0.976	0.627*	0.317	0.800*	0.265*
CLEF-2019-dta	7b-chat	Unc	0.088	0.924	0.676	0.340	0.375	0.416
		Cal (0.95)	0.087*	0.928*	0.664	0.323	0.500	0.388
		Cal (1)	0.082*	0.958*	0.626	0.312	0.625	0.283
	13b-chat	Unc	0.226	0.427	0.661	0.301	0.000	0.880
		Cal (0.95)	0.105	0.902	0.687	0.373	0.500*	0.451*
		Cal (1)	0.087*	0.962	0.617	0.326	0.750*	0.258*
CLEF-2019-Int	7b-chat	Unc	0.080	0.809	0.640	0.332	0.400	0.456
		Cal (0.95)	0.074*	0.921*	0.652*	0.335	0.486	0.369
		Cal (1)	0.068*	0.956*	0.625	0.327	0.649	0.282
	13b-chat	Unc	0.243	0.485	0.662	0.367	0.075	0.821
		Cal (0.95)	0.103*	0.914	0.732	0.390	0.514*	0.530*
		Cal (1)	0.084*	0.961	0.693	0.365	0.730*	0.407*