Chain of Ideas: Revolutionizing Research Via Novel Idea Development with LLM Agents

TL;DR

This work introduces Chain-of-Ideas (CoI), an LLM-driven agent that organizes literature into forward-backward chains to mirror the progressive development of a research field, enhancing idea generation. It couples CoI Construction with multi-branch exploration, novelty checking, and integrated experiment design, all evaluated through Idea Arena, an arena-style, pairwise evaluation aligning well with human judgments. Results show CoI outperforms existing automated baselines and approaches human performance in novelty and significance, while also delivering cost-efficient candidate ideas. The study further analyzes how CoI length and width affect performance, highlighting the value of structured, trend-aware literature organization for robust ideation.

Abstract

Effective research ideation is a critical step for scientific research. However, the exponential increase in scientific literature makes it challenging for researchers to stay current with recent advances and identify meaningful research directions. Recent developments in large language models~(LLMs) suggest a promising avenue for automating the generation of novel research ideas. However, existing methods for idea generation either trivially prompt LLMs or directly expose LLMs to extensive literature without indicating useful information. Inspired by the research process of human researchers, we propose a Chain-of-Ideas~(CoI) agent, an LLM-based agent that organizes relevant literature in a chain structure to effectively mirror the progressive development in a research domain. This organization facilitates LLMs to capture the current advancements in research, thereby enhancing their ideation capabilities. Furthermore, we propose Idea Arena, an evaluation protocol that can comprehensively evaluate idea generation methods from different perspectives, aligning closely with the preferences of human researchers. Experimental results indicate that the CoI agent consistently outperforms other methods and shows comparable quality as humans in research idea generation. Moreover, our CoI agent is budget-friendly, with a minimum cost of \$0.50 to generate a candidate idea and its corresponding experimental design.

Figures (7)

• Figure 1: Comparison between the vanilla retrieval augmented generation (RAG) research agent and our Chain-of-Ideas agent on the idea generation task.
• Figure 2: Our proposed CoI agent framework.
• Figure 3: Evaluation results of idea generation with LLM as a judge.
• Figure 4: Evaluation results of idea generation with human as judges.
• Figure 5: Agreements between human and LLM judges.