TaxCE : A Framework for Automated Taxonomy Construction and Evaluation at Scale
2026-08-31 • Computation and Language
Computation and Language
AI summaryⓘ
The authors developed TaxCE, a method that organizes large amounts of messy text feedback into clear, multi-level topic hierarchies. Unlike previous methods, TaxCE builds detailed topic structures from the ground up and checks its results using new measures that evaluate how complete, distinct, and detailed the topics are. They use these measures in a loop to keep improving the taxonomy until it fits the text well. Tests show TaxCE works better than older models in making useful and easy-to-navigate topic trees, and humans agree it produces higher-quality results.
hierarchical taxonomynatural language processingtopic modelingsemantic unitsexclusivityexhaustivitygranularitycorpus-grounded evaluationiterative refinementlarge language models
Authors
Sandeep Sricharan Mukku, Albert Aristotle Nanda, Rohit Pyati
Abstract
Organizing unstructured feedback text into hierarchical taxonomy is a fundamental challenge in NLP, particularly in domains where feedback arrives at massive scale in varied forms such as reviews, transcripts, and surveys. Existing approaches either produce shallow hierarchies, neglect long-tail topics, or lack rigorous evaluation frameworks. We present TaxCE, a fully automated framework that constructs multi-level hierarchical taxonomies from raw text through progressive condensation of corpus content into actionable segments, deduplicated semantic units, and granular topics with definitions, which are then organized bottom-up into a hierarchy with corpus-groundedness. We also introduce three corpus-grounded evaluation metrics, Exclusivity, Exhaustivity, and Granularity (EEG), and integrate them into a metrics-in-the-loop iterative refinement mechanism that diagnoses deficiencies and applies targeted corrections until convergence. Extensive experiments demonstrate that TaxCE consistently outperforms existing baselines spanning classical topic models, neural methods, and LLM-based approaches, with average improvements of 11.8, 20.5, and 15.7 percentage points in exclusivity, exhaustivity, and granularity respectively over the strongest baseline. Human evaluation further confirms superior taxonomy quality, actionability, and navigability.