AI-powered LLM taxonomy tool enhances research efficiency

AI researchers now have a new tool to navigate the complex landscape of AI safety research papers. TRecursive, a project developed by Myles H, uses LLMs to generate hierarchical taxonomies from research paper collections, providing an interactive visual map of academic fields. The system has been tested on over 3,000 AI safety papers from ArXiv, creating a navigable structure that helps researchers gain perspective on how individual papers fit into broader research contexts.

The big picture: TRecursive combines automated taxonomy generation with an intuitive visualization interface to make large collections of research papers more accessible and interconnected.

• The system recursively categorizes papers into a hierarchical structure, continuing to subdivide categories until reaching manageable leaf nodes.
• The resulting taxonomy is displayed as an interactive map at trecursive.com/ai-safety-taxonomy, allowing researchers to explore both high-level research domains and specific paper collections.

How it works: The taxonomy generation process uses a series of LLM operations to create, evaluate, and refine category structures for research papers.

• An initial LLM proposes category breakdowns for a given research field, which is then tested by sorting paper samples into these categories.
• The system evaluates each proposed taxonomy using five metrics, including how exclusively papers fit into single categories and how evenly papers distribute across categories.
• After several iterations, the highest-scoring taxonomy is selected, and all papers are sorted accordingly before the process repeats for each subcategory.

Key features: The TRecursive interface offers two main viewing modes designed for different exploration needs.

• “Titles mode” provides a bird’s-eye view of the research landscape, with hover functionality revealing brief descriptions and related topics.
• “Default mode” allows users to dive deeper into specific topics and view all papers sorted within a particular category.

Why this matters: Academic research fields like AI safety are growing rapidly, making it increasingly difficult for researchers to maintain a comprehensive understanding of the landscape.

• TRecursive aims to reduce information overload by providing structured maps that highlight connections between research areas.
• The tool could potentially help researchers identify gaps in the literature, find relevant prior work, and understand how their contributions fit into broader research contexts.

Behind the numbers: The evaluation system uses five weighted metrics to select the most effective taxonomies.

• The system prioritizes taxonomies where papers fit cleanly into single categories (rather than spanning many) and maximizes the percentage of papers that can be placed within the taxonomy.
• Additional scoring factors include feedback from multiple LLMs, the presence of overview papers in categories, and how evenly papers distribute across categories.