Link: http://arxiv.org/abs/2506.02426v1
PDF Link: http://arxiv.org/pdf/2506.02426v1
Summary: Entity relationship classification remains a challenging task in informationextraction, especially in scenarios with limited labeled data and complexrelational structures.
In this study, we conduct a comparative analysis ofthree distinct AI agent architectures designed to perform relationclassification using large language models (LLMs).
The agentic architecturesexplored include (1) reflective self-evaluation, (2) hierarchical taskdecomposition, and (3) a novel multi-agent dynamic example generationmechanism, each leveraging different modes of reasoning and prompt adaptation.
In particular, our dynamic example generation approach introduces real-timecooperative and adversarial prompting.
We systematically compare theirperformance across multiple domains and model backends.
Our experimentsdemonstrate that multi-agent coordination consistently outperforms standardfew-shot prompting and approaches the performance of fine-tuned models.
Thesefindings offer practical guidance for the design of modular, generalizableLLM-based systems for structured relation extraction.
The source codes anddataset are available at\href{https://github.
com/maryambrj/ALIEN.
git}{https://github.
com/maryambrj/ALIEN.
git}.
Published on arXiv on: 2025-06-03T04:19:47Z