Data and Resources

Claude Code Plugins

• Agentic clustering Discover the natural groupings in a collection of texts, then sort every text into them. Install instructions Paper

  Give it a corpus, tell it what you care about (for example “group these articles by the policy issue they discuss”), and set roughly how many clusters you want. A team of AI agents then builds the taxonomy for you: one proposes clusters, one synthesises them, one audits them, one investigates unclear cases and one critiques the result. An orchestrator steers the whole loop, and it runs unattended until the taxonomy stops changing. You get a readable taxonomy.md with definitions for each cluster.

  A second command classifies your whole corpus into those clusters, giving a label, a confidence score and reasoning for each text. That step can use an LLM, or a small model you train and run on your own machine for free. On seven public text-clustering benchmarks it is state-of-the-art, beating the strongest prior method by 47% in mean ARI.

  To install, run inside Claude Code:

  /plugin marketplace add emilysilcock/econ-nlp-plugins
  /plugin install agentic-clustering@econ-nlp-plugins

Datasets

• Daily Disability-Related Hate Crimes Recorded by Police Forces in England and Wales, 2013–2023 (Freedom of Information Returns) doi:10.7910/DVN/KVHQ43 Harvard Dataverse
• Daily Alleged Benefit Fraud Reports Recorded by the Department for Work and Pensions in Great Britain and England, 2013–2023 (Freedom of Information Returns) doi:10.7910/DVN/YNQYKK Harvard Dataverse