EvolveSearch: An Iterative Self-Evolving Search Agent

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhang, Dingchu, Zhao, Yida, Wu, Jialong, Li, Baixuan, Yin, Wenbiao, Zhang, Liwen, Jiang, Yong, Li, Yufeng, Tu, Kewei, Xie, Pengjun, Huang, Fei
Format:	Preprint
Published:	2025
Subjects:	Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866909626174275584
author	Zhang, Dingchu Zhao, Yida Wu, Jialong Li, Baixuan Yin, Wenbiao Zhang, Liwen Jiang, Yong Li, Yufeng Tu, Kewei Xie, Pengjun Huang, Fei
author_facet	Zhang, Dingchu Zhao, Yida Wu, Jialong Li, Baixuan Yin, Wenbiao Zhang, Liwen Jiang, Yong Li, Yufeng Tu, Kewei Xie, Pengjun Huang, Fei
contents	The rapid advancement of large language models (LLMs) has transformed the landscape of agentic information seeking capabilities through the integration of tools such as search engines and web browsers. However, current mainstream approaches for enabling LLM web search proficiency face significant challenges: supervised fine-tuning struggles with data production in open-search domains, while RL converges quickly, limiting their data utilization efficiency. To address these issues, we propose EvolveSearch, a novel iterative self-evolution framework that combines SFT and RL to enhance agentic web search capabilities without any external human-annotated reasoning data. Extensive experiments on seven multi-hop question-answering (MHQA) benchmarks demonstrate that EvolveSearch consistently improves performance across iterations, ultimately achieving an average improvement of 4.7\% over the current state-of-the-art across seven benchmarks, opening the door to self-evolution agentic capabilities in open web search domains.
format	Preprint
id	arxiv_https___arxiv_org_abs_2505_22501
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	EvolveSearch: An Iterative Self-Evolving Search Agent Zhang, Dingchu Zhao, Yida Wu, Jialong Li, Baixuan Yin, Wenbiao Zhang, Liwen Jiang, Yong Li, Yufeng Tu, Kewei Xie, Pengjun Huang, Fei Computation and Language The rapid advancement of large language models (LLMs) has transformed the landscape of agentic information seeking capabilities through the integration of tools such as search engines and web browsers. However, current mainstream approaches for enabling LLM web search proficiency face significant challenges: supervised fine-tuning struggles with data production in open-search domains, while RL converges quickly, limiting their data utilization efficiency. To address these issues, we propose EvolveSearch, a novel iterative self-evolution framework that combines SFT and RL to enhance agentic web search capabilities without any external human-annotated reasoning data. Extensive experiments on seven multi-hop question-answering (MHQA) benchmarks demonstrate that EvolveSearch consistently improves performance across iterations, ultimately achieving an average improvement of 4.7\% over the current state-of-the-art across seven benchmarks, opening the door to self-evolution agentic capabilities in open web search domains.
title	EvolveSearch: An Iterative Self-Evolving Search Agent
topic	Computation and Language
url	https://arxiv.org/abs/2505.22501