Organization of Power

Literature Review II: Article Corpus to Arguments

Overview

Video not available yet

Estimated duration: 120 minutes

Overview

Four chapters on using AI to turn a corpus of academic articles into a map of the arguments.

Scope

Four chapters on turning a corpus of academic articles into a map of the arguments and a place where your own argument can enter.

Series

Part 2 of 3 · AI x Literature Review

Format

Tutorial · 4 chapters · about 2 hours

Worked example

Post-socialist transformation: a 434-work corpus on North Korea, China, and Vietnam

Getting from a pile of relevant articles to an organized set of arguments is the step graduate students most often skip, and the one where AI changes the economics most: a model can read three hundred articles overnight, but it cannot tell you which question is worth asking, and it will confidently summarize an article it has only seen the abstract of.

The tutorial breaks the step into four moves, one per chapter, each producing a file the next one reads. You explore the corpus without a question (Chapter 1), query it with a research question and sort the works into a core and a periphery (Chapter 2), organize the core arguments into a ledger and a typology (Chapter 3), and find your entry point among the untested scope conditions and unresolved disagreements (Chapter 4).

Every prompt was run on a real corpus and every output shown is real: 434 works on the political economy of reform in North Korea, China, and Vietnam, 309 of them with a full text. Chapters 2 to 4 follow one question — how socialist and post-socialist centres direct attention and resources to particular enterprises and localities — which connects this tutorial to its companion, Generating Data I: Extracting Datasets from a Large Media Corpus, where a leader's on-site visit becomes an observable measure of that attention.

This is Part 2 of the AI x Literature Review series. Part 1, From Seeds to Corpus, builds the corpus, and Part 3, From Arguments to Prose, writes the review; both are in preparation.

What you need

  • A corpus exported to plain files. The lab exports each corpus as citation_index.csv (one row per work, keyed by its Zotero item_key), a text/ folder of full texts named by item key, and scores.csv from Part 1. The item key ties every claim back to a PDF you can open.
  • An AI assistant. A local agent that can open files (Claude Code, Codex) can read every full text and save its results; it suits the survey and the research-question analysis, where coverage matters. A web chat with uploads takes the citation index whole but full texts only in batches of about 50; it suits a first look from abstracts or a close reading of a few core articles. The prompts are written for a local agent but work in a web chat if you supply the files they name.
  • The PDFs, in Zotero, for spot-checking what the AI says about them.

How to read this

The chapters are best read in order: each step uses the output of the one before. If you already have a research question you can start at Chapter 2, but Chapter 1 is where you find out whether your question is one the field already asks.

Five ground rules are written into every prompt, because an AI that is not told them will not follow them: cite every article by title and item key; never invent an article; say when only the abstract was used; quote, for anything that matters; and spot-check five of the cited articles against their PDFs before relying on any output. Each chapter closes with the opportunities and the challenges of using AI at that step.

Passages marked IN PRACTICE, with a blue rule down the left, give the exact files, prompts, and settings. Passages marked in amber are cautions: the places where an AI output looked right and was not.

Content

Part 2 of the AI x Literature Review series. Read the four chapters in order. Each chapter gives the prompt, the real output from the worked example, and the opportunities and challenges of using AI at that step.

Chapters 1–4 From article corpus to arguments Each chapter is one page. Read them in order the first time.
  1. 01 Exploring the CorpusCounts by decade, clusters by discipline and topic, and a country cross-tab. What an unsupervised survey shows and how to check it.
  2. 02 Asking a Research QuestionHow do socialist and post-socialist centres direct attention to particular enterprises and localities? One question run against 434 works, with the variations, arguments, data, and methods of the core and the periphery.
  3. 03 Organizing the ArgumentsAn arguments ledger for the core articles, each argument on three levels, and a typology that shows who disagrees with whom.
  4. 04 Finding Your Entry PointTwo kinds of gap — an untested scope condition or factor, and an unresolved disagreement — as candidate contributions, and why the choice among them is yours.

Workshops

Resources

Relevant Literature

Browse references and supporting literature for this training module in the Organization of Power Lab Zotero library.

Open this training module's Zotero collection

A note on the example. The corpus was built for the lab's research on subnational economic management in North Korea. It holds published works only. Article titles and Zotero item keys are real; the full texts are licensed and are not redistributed here.

A note on the outputs. The analyses in these chapters were produced by an AI agent working on the lab's copy of the corpus in October 2026 and then checked. Where a check found a problem, the chapter says so. Treat the outputs as a demonstration of the method, not as a finished review of the field.

Organization of Power Lab · National Chengchi University, Taipei. Written 2026.