notesum.ai

Published at October 30

Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval

cs.AI

Released Date: October 30, 2024

Authors: Le Huang¹, Hengzhi Lan¹, Zijun Sun², Chuan Shi¹, Ting Bai¹

Aff.: ¹Beijing University of Posts and Telecommunications; ²Yunic.AI

Arxiv: http://arxiv.org/abs/2410.23041v1

Refer to caption

Agent Types	Methods	BFI				MBTI
Agent Types	Methods	Acc(Dim) $\uparrow$	Acc(Full) $\uparrow$	MSE $\downarrow$	MAE $\downarrow$	Acc(Dim) $\uparrow$	Acc(Full) $\uparrow$	MSE $\downarrow$	MAE $\downarrow$
ChatGLM-6B	Ordinary RAG	0.6242	0.1250	0.1849	0.3728	0.6694	0.2188	0.1526	0.3610
ChatGLM-6B	Emotional RAG	0.6369	0.0938	0.1720	0.3625	0.6694	0.2812	0.1539	0.3655
Qwen-72B	Ordinary RAG	0.6815	0.0938	0.1433	0.3024	0.7438	0.3438	0.1230	0.2920
Qwen-72B	Emotional RAG	0.7261	0.2500	0.1269	0.2878	0.7934	0.4688	0.1156	0.2900
GPT-3.5	Ordinary RAG	0.7006	0.1875	0.1496	0.3121	0.7851	0.5000	0.1221	0.2965
GPT-3.5	Emotional RAG	0.7006	0.1875	0.1475	0.3082	0.7851	0.4375	0.1236	0.2927