Step 2 · Search Prep

Let the seeds propose the search terms, choose a short and balanced set, decide which journals each search covers, and translate the terms for the databases the agent cannot search.

Part 1 · Step 2 of Steps 0–5
Output · 40 approved search terms, four journal lists and translated search strings

A search is only as good as its terms and its sources. Here the agent does the counting, proposing terms that the seeds use often and that distinguish them from ordinary academic prose, and the researcher does the choosing. The aim is a short list that covers the review's concepts without letting any one of them dominate.

Section 1Candidate terms from the seeds

The agent extracted single words and two- and three-word phrases from the seed texts and from each reference document, ranked them, and merged the three rankings with the weights set in Step 1, one half for the seeds and one quarter for each reference document. The merged list held 49,794 candidate terms. Most are noise, such as common academic wording or place names that every article mentions, and the agent prepared a shortlist for the researcher.

Section 2Choosing the terms

The review's configuration asked for a portfolio of ten general terms, two single words, six two-word phrases and two three-word phrases. Single words find a great deal and much of it is irrelevant. Long phrases are precise but miss works that say the same thing differently. The researcher chose the ten terms from the shortlist.

Table 2.1The ten general search terms
LengthTerms
One wordmarketization · decentralization
Two wordseconomic reform · fiscal decentralization · local governments · enterprise management · transition economies · economic transformation
Three wordspost-socialist transformation · local state corporatism
From 02_derive-search-terms, approved by the researcher.

Section 3Country sets

For the three area searches the agent prefixed each general term with a country name, as in North Korea marketization or Vietnam fiscal decentralization. That gave three country sets of ten and, with the general set, 40 terms in all. A term that the country name makes too narrow can be replaced in one set without touching the others.

Fewer terms in later reviews

After this review the lab's skill was revised to use five general terms and five per country. Every term is searched in every journal on the list, so each term removed saves as many queries as there are journals.

Section 4Journal lists

Each search runs over a list of journals as well as over two open indexes, Crossref and OpenAlex, that cover the whole academic literature. The journal lists keep the searches close to the fields that matter. For each search the agent compiled candidate lists and the researcher chose among them before approving. For North Korea, Vietnam and the general search the researcher asked for new lists, and the China search drew on an existing list of Chinese social science journals.

Table 2.2Journals selected for each search
SearchJournalsTerms
General4210
North Korea7710
China11710
Vietnam6210
The four lists are the journals each search actually queried in Step 3.

Section 5Terms for the special platforms

Much of the literature on China and North Korea is published in Chinese and Korean journals indexed by CNKI, Airiti, KCI and RISS. These databases do not offer an open interface for automated searching, and their licences forbid scraping, so they are searched by a person. The agent's part is to translate each term into the language and syntax of each database and to write a capture guide that says exactly what to type, how many results to save and where to save them in Zotero.

Table 2.3Three terms as written for each database
TermCNKI (simplified)Airiti (traditional)KCI and RISS (Korean)
marketizationSU='市场化'市場化북한 시장화
fiscal decentralizationSU=('财政分权'+'财政包干'+'分灶吃饭')財政分權 OR 財政包乾북한 지방예산제 OR 북한 재정분권
transition economiesSU=('转型经济'+'转轨经济')轉型經濟 OR 轉軌經濟체제전환국 OR 체제전환 경제
From the draft translations in 03_search-journals. The Korean strings add the prefix 북한 (North Korea) where the term would otherwise return South Korean work, and add local terms such as 지방예산제, the North Korean local budget system.

The capture guide for Airiti is reproduced, in English and with its links made inactive, on the example Airiti page.

Translation is a judgment

A literal translation can miss the term the field actually uses. The Chinese literature on fiscal reform speaks of 财政包干 (fiscal contracting) more than of fiscal decentralization, and North Korean studies in Korean use their own vocabulary. The researcher, or someone who reads the language, should check every string before it is searched.

  • search termsThe candidate rankings, the shortlist and the 40 approved terms.02_derive-search-terms
  • journal listsOne approved list per search.03_search-journals
  • translated_terms_draft.csvEach general term in CNKI, Airiti and Korean syntax, with notes.03_search-journals

Terminology

Candidate term
A word or phrase proposed by the agent from the seed texts, before the researcher's selection.
Term portfolio
The planned mix of one-, two- and three-word terms, balancing how much a search finds against how precise it is.
Country set
The general terms with a country name attached, used for one area search.
Open index
A free bibliographic database covering many publishers, here Crossref and OpenAlex, that an agent can query directly.
Capture guide
A page written by the agent telling a person what to search on a special platform and how to save the results to Zotero.