notesum.ai

Published at October 22

Literature Meets Data: A Synergistic Approach to Hypothesis Generation

cs.LG

cs.AI

Released Date: October 22, 2024

Authors: Haokun Liu¹, Yangqiaoyu Zhou¹, Mingxuan Li¹, Chenfei Yuan², Chenhao Tan¹

Aff.: ¹University of Chicago; ²Tsinghua University

Arxiv: https://arxiv.org/abs/2410.17309v1

Refer to caption

GPT-4 MINI	No hypothesis
Model	Methods	Deceptive reviews	LlamaGC	GPTGC	Persuasive Pairs	Dreaddit
	Zero-shot	55.47	50.00	56.33	81.24	64.60
	Few-shot k=3	65.56	51.11	64.22	83.64	75.00
	Zero-shot generation	68.69	49.00	53.00	86.08	65.00
	Literature-based
	literature-only	59.22	49.00	54.00	78.80	67.68
	HyperWrite	61.63	49.67	52.67	82.36	68.76
	NotebookLM	53.03	49.33	51.67	68.96	62.28
	Data-driven
	HypoGeniC	75.22	81.67	68.56	82.20	76.56
	Literature + Data (This work)
	HypoRefine	77.78	55.33	63.33	89.04	78.04
	Literature $\cup$ HypoGeniC	72.41	83.00	69.22	89.88	78.20
	Literature $\cup$ HypoRefine	77.19	55.33	63.00	89.52	79.24
Llama 70B-I	No hypothesis
	Zero-shot	62.87	58.67	63.00	85.60	64.56
	Few-shot k=3	68.56	70.45	76.00	86.80	69.44
	Zero-shot generation	56.28	50.67	55.67	88.16	66.16
	Literature-based
	literature-only	64.25	50.00	49.67	80.56	66.04
	HyperWrite	58.62	50.67	54.00	83.24	74.40
	NotebookLM	57.81	49.33	50.67	67.64	66.56
	Data-driven
	HypoGeniC	62.06	78.67	78.00	88.44	75.48
	Literature + Data (This work)
	HypoRefine	72.16	67.00	66.67	87.52	78.92
	Literature $\cup$ HypoGeniC	73.72	81.33	78.67	86.72	72.56
	Literature $\cup$ HypoRefine	71.75	66.67	65.67	88.76	74.80