Researchers have developed EvoTree, a novel framework designed to automatically generate evolution trees from citation graphs, addressing the limitations of traditional surveys and existing taxonomy-induction methods. EvoTree decouples backbone learning from temporal refinement, using a graph-aware encoder and hierarchical clustering for a stable taxonomy, followed by temporal fine-tuning and an LLM pass for concept labeling. The framework demonstrates superior performance in concept purity and accuracy on a new benchmark dataset across 11 AI subfields, outperforming existing methods in detecting marginal papers and maintaining topological consistency. AI
IMPACT Automates the creation of research lineage maps, potentially accelerating knowledge discovery and synthesis in AI subfields.
RANK_REASON The item describes a new research paper introducing a novel framework for academic research. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX Code Finder for Papers
- Connected Papers
- CORE Recommender
- DagsHub
- EvoTree
- Gotit.pub
- Hugging Face
- Influence Flower
- Litmaps
- ScienceCast
- scite Smart Citations
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
