EvoTaxo: Building and Evolving Taxonomy from Social Media Streams
arXiv cs.CL / 3/23/2026
📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research
Key Points
- EvoTaxo is an LLM-based framework for building and evolving taxonomies from temporally ordered social media streams.
- It converts each post into a structured draft action over the current taxonomy rather than clustering raw posts.
- The approach accumulates structural evidence over time windows and uses dual-view clustering that combines semantic similarity with temporal locality, followed by refinement/arbitration to select reliable edits.
- Each taxonomy node maintains a concept memory bank to preserve semantic boundaries over time, supporting evolving taxonomies.
- Experiments on Reddit datasets show more balanced taxonomies, clearer post-to-leaf assignment, better corpus coverage at comparable taxonomy size, and stronger structural quality, with a case study illustrating meaningful temporal shifts; the codebase is available.
Related Articles
The Moonwell Oracle Exploit: How AI-Assisted 'Vibe Coding' Turned cbETH Into a $1.12 Token and Cost $1.78M
Dev.to
How CVE-2026-25253 exposed every OpenClaw user to RCE — and how to fix it in one command
Dev.to
Day 10: An AI Agent's Revenue Report — $29, 25 Products, 160 Tweets
Dev.to
Does Synthetic Data Generation of LLMs Help Clinical Text Mining?
Dev.to
What CVE-2026-25253 Taught Me About Building Safe AI Assistants
Dev.to