MAPEX: A Multi-Agent Pipeline for Keyphrase Extraction Using LLMs

Keyphrase extraction is a fundamental task in natural language processing. However, existing unsupervised prompt-based methods with Large Language Models (LLMs) typically rely on simplistic prompt designs, which limit their robustness across document lengths and hinder the full exploitation of LLMs’ reasoning and generation capabilities. To address these challenges, we propose MAPEX, the first framework that introduces multi-agent collaboration into keyphrase extraction. MAPEX coordinates an LLM-based Extractor and an Enhancer, leveraging modules for role-based prompting, topic guidance, knowledge augmentation, and post-processing. A dual-path strategy enables dynamic adaptation to document length: short texts are processed via knowledge-driven extraction, while long texts benefit from topic-guided extraction. Extensive experiments on six benchmark datasets across three different LLMs demonstrate its strong generalization and universality, outperforming the state-of-the-art (SOTA) unsupervised method by 2.44% and standard LLM baselines by 4.01% in F1@5 on average. Our code is publicly available at https://github.com/NKU-LITI/MAPEX.