DR − CoT: A Multi-Agent Framework for Enhanced Reasoning in Large Language Models

In the world of artificial intelligence (AI), the ability of large language models (LLMs) to reason through natural language has become a game-changer. This ability allows AI systems to understand complex questions, make decisions, and extract useful insights from vast amounts of data, transforming industries like customer service, education, and healthcare. One of the toughest challenges is enabling AI systems to answer multi-hop questions, which require pulling information from multiple sources to connect the pulled information and provide an accurate answer. Answering such questions, involve relationships between different pieces of data is far more complicated than simply retrieving a single fact of one-hop query. Despite significant progress, there are still hurdles in getting AI to balance how it retrieves information, reasons through it, and generates accurate answers. To tackle these challenges, we introduce DR-CoT, a dynamic rephrasing-based multi-agent framework designed to boost the reasoning abilities of LLMs in answering complex, multi-hop questions. Our approach breaks down difficult queries into simpler, step-by-step sub-questions, refining them along the way with targeted keyword searches and validation to fill in any missing links. This method ensures a smooth, efficient, and more accurate reasoning process, making it easier for AI to handle complex tasks. DR-CoT offers a promising path toward more intuitive, effective, and scalable AI systems that can reason like humans. Experiments on benchmark datasets demonstrate significant performance gains, with improved answer accuracy and robustness for real-world applications. This research advances intelligent QA systems, paving the way for reliable AI in fields like education and healthcare, while addressing retrieval errors and strengthening capabilities for complex reasoning processes.