Deep learning–based knowledge tracing, 2015–2025: derived bibliometric and taxonomy tables (Scientometrics submission)

Le résumé fourni par la source

science_mapping/vosviewer_export_n926/vos_keywords_n926_{map,network}.txt` and `vos_countries_n926_{map,network}.txt` are replaced with genuine VOSviewer 1.6.20output. **v4.1.0–v4.1.2 shipped files that were formatted for VOSviewer but whose item coordinates and cluster assignments were computed by NetworkX (spring_layout + Louvain), not by VOSviewer itself, despite manuscript captions stating otherwise.** This was caught, and is corrected here: the map files were stripped to id/label/score only, reopened in the VOSviewer 1.6.20 GUI (Analysis tab: association-strength normalization; keywords at VOSviewer defaults — resolution 1.00, minimum cluster size 1; countries with Layout repulsion raised from the default 1 to 5, since several near-isolated country pairs otherwise resolved to coincident coordinates), and the map/network VOSviewer itself wrote back out. Verified against the old NetworkX baseline with `vosviewer/n926/verify_vos_output.py` (coordinates and cluster partition structure both differ from the NetworkX version; item set, link columns, and year score unchanged) before this package was built. Keyword clusters: 13 (was 11 under the mislabeled Louvain partition). Country clusters: 5 (same partition structure as before; explained in the SI Fig. S10 caption as expected given how sparse/disconnected the 17-country graph is, not evidence the files were not recomputed — country coordinates changed substantially even though the partition did not). Record-level N=926 codes, A3 reindex (N_U=335), and the A1 blind-recode package (seed 20260817, n=60) are **unchanged** from v4.1.2.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.