Screening data for a systematic mapping of Scopus-indexed 2026 literature on artificial intelligence in Chinese education (SAMYRAD 2026, contribution 302)

Abstract

Record-level screening data of a systematic mapping of the 2026 literature on artificial intelligence in Chinese education, companion to the paper "Digital transformation and systemic architecture of Chinese education: trajectory and the AI + Education plan toward 2030" (SAMYRAD 2026, Seville, 5–6 October 2026, IEEE Xplore proceedings). The corpus was retrieved from Scopus on 19 May 2026 with the search string TITLE-ABS-KEY (China AND education AND "artificial intelligence") AND PUBYEAR = 2026 (N = 363 records: 265 articles, 40 conference papers, 24 reviews, 18 book chapters, 16 other document types; 35 in press). Records were screened at title and abstract level against two inclusion criteria (setting in the Chinese education system; at least one pedagogical, attitudinal or competency outcome) and four hierarchically applied exclusion criteria (document type; AI not the object or outside an instructional setting; no primary outcome data; non-Chinese or unspecified setting), the first applicable criterion being the one recorded. Screening was performed in two passes. In the first pass, carried out on 6 September 2026, a large language model (Claude, Anthropic) applied the criteria to each title and abstract and recorded a proposed decision with its criterion, its sub-reason and a borderline flag. In the second pass, one author examined all 363 records with the model output visible and recorded the decision that prevails in every count reported in the companion paper. Because the second pass was not blinded to the first, the agreement figures in the Summary sheet are descriptive only; they are not an inter-rater reliability estimate, and no reliability coefficient is reported in the paper. Retained records were assigned a primary and, where relevant, a secondary theme against a codebook (acceptance models and cognitive friction; socio-territorial stratification; AI literacy frameworks; disciplinary specialisation and local language models; and an open category). Known limits of the first pass: abstracts only, occasionally truncated on reading; no affiliation field in the export; no full-text check. Files: search protocol (query, date, export settings, record composition); screening workbook with one row per record (identifiers, title, document type, publication stage, the proposed decision, criterion and sub-reason of the first pass, the author decision, theme and comment of the second pass, primary and secondary themes, borderline flags), a Summary sheet (decision and theme counts for both passes and descriptive agreement between them) and a README sheet with the codebook; a CSV of record identifiers (Scopus ID, EID, DOI); Scopus advanced-search queries that allow the corpus to be reconstructed from those identifiers; and the codebook as a separate file.