Researchers Develop Hierarchical Taxonomy Framework to Track AI Trends
Sørbø and his team have developed a new methodology based on a hierarchical taxonomy to analyze artificial intelligence trends from diverse sources such as arXiv, Hugging Face, and Deep Learning Weekly.
A new study published in the journal EPJ Data Science presents an analytical framework that integrates different data sources into a common hierarchical structure to monitor technical developments in the field of artificial intelligence.
The method developed by Sørbø and his colleagues allows for multi-resolution comparative analysis by mapping AI-related texts at different conceptual levels. Using directed acyclic graphs (DAGs), the system organizes broad categories at top nodes and specialized subfields at lower nodes.
The study employs five key metrics: topic popularity, depth of detail, diversity, temporal uniformity, and temporal uniqueness. The researchers analyzed data from arXiv preprints, Hugging Face model cards, and the Deep Learning Weekly newsletter using the LLMDap pipeline and the GPT-4.1-nano model.
The paper titled 'Multi-source, multi-level: a hierarchical taxonomy-based framework for AI trends analysis,' published in 2026, provides a scientific tool for tracking technical shifts in the world of artificial intelligence.
Sources
- bioengineer.org · Link