Enhancing Research Idea Generation through Combinatorial Innovation and Multi-Agent Iterative Search Strategies

Scientific progress relies on the continuous emergence of innovative discoveries. However, the exponential growth in scientific literature has increased the cost of information filtering, making it significantly more challenging for scientists to identify innovative research directions. Although artificial intelligence (AI) methods have shown potential in tasks such as research idea generation and hypothesis formulation, the ideas they produce are often repetitive and simplistic. Combinatorial innovation theory posits that new entities arise from the recombination of existing elements, offering a novel approach to addressing these challenges. This study draws on combinatorial innovation theory and the Delphi method to introduce a multiagent iterative planning and search strategy into the research idea generation process, aiming to enhance the diversity and novelty of generated ideas. The strategy integrates iterative knowledge search with a large language model (LLM)-based multi-agent system to iteratively generate, evaluate, and refine research ideas. Experiments conducted using data from the field of natural language processing demonstrate that the multi-agent iterative planning and search strategy outperforms stateof-the-art methods in terms of diversity and novelty, showcasing its potential to generate high -quality research ideas.This study not only validates the effectiveness of the multi-agent iterative search strategy but also provides a theoretical explanation, grounded in combinatorial innovation theory and methodologies, for its ability to improve research idea generation performance. It offers new perspectives for future work in this domain.