notesum.ai

Published at October 23

An Adaptive Framework for Generating Systematic Explanatory Answer in Online Q&A Platforms

cs.LG

cs.AI

cs.CL

Released Date: October 23, 2024

Authors: Ziyang Chen¹, Xiaobin Wang², Yong Jiang², Jinzhi Liao¹, Pengjun Xie², Fei Huang², Xiang Zhao¹

Aff.: ¹Independent Researcher, Changsha, China; ²Alibaba Group, Hangzhou, China

Arxiv: https://arxiv.org/abs/2410.17694v1

Refer to caption

Model	LLM-base Score						Information Score
Model	Overall	Fluency	Relevance	Logic	Reference	Depth	Overall (%)
ChatGPT	3.76	3.92	4.34	4.00	3.50	3.24	\
GPT-4	3.82	3.90	4.40	3.90	3.64	3.28	\
Qwen-Max	4.06	4.00	4.58	4.28	3.96	3.88	63.91
RAG	4.12	3.94	4.70	4.22	4.04	3.86	66.43
OutlineRAG	4.34	4.00	4.60	4.22	4.16	4.32	83.01
SynthRAG	4.52	3.96	4.74	4.30	4.40	4.54	85.76
w/o generation	3.86	4.00	4.18	3.76	3.78	3.58	74.44
w/o customized	4.42	3.96	4.66	4.18	4.14	4.28	85.64