Towards Automatic Evolution Tree Generation from Citation Graphs

Researchers proposed EvoTree, a staged framework that generates evolution trees from citation graphs by decoupling conceptual backbone learning from temporal refinement, according to the arXiv paper 2609.09561v1. EvoTree combines a graph-aware encoder with distribution-based hierarchical clustering for a stable taxonomy backbone, then applies temporal fine-tuning to re-attach marginal papers to internal nodes under monotonic-path constraints, with a final LLM pass labeling concepts without altering topology. The authors released the first annotated benchmark for this task across 11 AI subfields, and report EvoTree attains the highest NMI and citation-direction accuracy among all baselines plus the best concept purity, and is the only method with non-trivial marginal-paper detection on the annotated set.

arXiv:2609.09561v1 Announce Type: new Abstract: Surveys remain the primary way researchers grasp the lineage of methods within an AI subfield, but they scale poorly against the current rate of publication. Existing taxonomy-induction methods are largely leaf-bound and time-agnostic; they tend to force transitional papers into mature leaves and can create topological inversions between ancestors and descendants. We propose EvoTree, a staged framework that decouples conceptual backbone learning from temporal refinement: a graph-aware encoder with distribution-based hierarchical clustering yields a stable taxonomy backbone; temporal fine-tuning then re-attaches marginal papers to internal nodes under monotonic-path constraints; a final LLM pass labels concepts without altering the topology. We release the first annotated benchmark for this task across 11 AI subfields. EvoTree attains the highest NMI and citation-direction accuracy among all baselines and the best concept purity on the annotated benchmark, and is the only method with non-trivial marginal-paper detection on the annotated set.