Deep learning–based knowledge tracing, 2015–2025: derived bibliometric and taxonomy tables (Scientometrics submission)
Rattachement africain : vn. Niveau de preuve : code pays fourni par la source.
Le résumé fourni par la source
Derived analysis tables for a locked Scopus-core corpus of deep learning–based knowledge tracing (DLKT) publications, 2015–2025 (N = 926; eligibility identity 1287 = 926 + 361 + 0). This deposit supports a Scientometrics manuscript on residual-aware science mapping: TITLE-ABS-KW taxonomy codes, bibliometric aggregates, exploratory association-strength networks, a stratified full-text residual audit (A3), and inter-rater reliability (IRR) on designated dual-coder overlap samples. **Version 4.1.4 increment (2026-08-22).** Adds full reproducibility for Supplementary Information per-label Cohen's κ tables on the multi-label `architecture_families` and `secondary_mechanisms` axes: - `irr_overlap/irr_overlap_n40_human_dual_labels_pre_adjudication.csv` — complete n = 40 pre-adjudication dual labels on the A2-NEW-145 human-only overlap (seed 20260813; v4.1.3 deposited agreement summaries and disagreements only).- `irr_overlap/per_label_kappa_architecture_families.csv` and `per_label_kappa_secondary_mechanisms.csv` — derived per-label reliability outputs.- `analysis_scripts/compute_per_label_kappa.py` — regeneration script (bootstrap B = 2000, seed 20260728; not estimable when n_pos < 3).- `analysis_scripts/build_n40_dual_labels.py` — optional utility to rebuild the n = 40 dual file from project coder returns. All v4.1.3 content is unchanged, including record-level N = 926 taxonomy codes, genuine VOSviewer 1.6.20 science-mapping export files, A3 Horvitz–Thompson / Hájek reindex on the current UNCERTAIN residual (N_U = 335), association-strength network tables, and the A1 step-2 blind NL2 re-coding package (seed 20260817; n = 60). **Package contents (full zip).** - Record-level TITLE-ABS-KW taxonomy codes and frequency / cross-tab tables (`l2_taxonomy/`).- De-identified pre-adjudication dual-coder labels: n = 94 overlap on the prior NL2 production subset; n = 40 human-only overlap on newly eligible records (`irr_overlap/`).- A3 residual-validation consensus and sensitivity tables; current-residual reindex (`a3_residual_validation/`, `review_reindex_n335/`).- Keyword thesaurus, network node/edge lists, parameter manifests, and VOSviewer map/network files (`networks/`, `science_mapping/`).- Research-agenda matrix, RQ3 pilot aggregates, classical bibliometric tables, and analysis-script pointers (`analysis_scripts/`). **Not included:** licensed Scopus raw exports, copyrighted full-text PDFs, or coder working sheets beyond the de-identified dual-label files above. **Claim boundaries:** TITLE-ABS-KW codes are not a full-text census of N = 926; A3 is a stratified audit (n = 194 AVAILABLE of 200 drawn), reindexed rather than redrawn; network layers are associative diagnostics only. **Versioning:** cite this version DOI for v4.1.4. Concept / all-versions DOI: 10.5281/zenodo.21635880. Prior version: v4.1.3 (10.5281/zenodo.21986524).
Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.
Le contrôle bibliographique ouvert
Où se fait cette recherche
-
Hung Yen University of Technology and Education pays non établi dans la noticeUniversité ou école supérieure
-
VNU University of Science pays non établi dans la noticeUniversité ou école supérieure
-
AIRCAAI / ITI VNU Hanoi pays non établi dans la noticeInstitution
Hung Yen University of Technology and Education, VNU University of Science et AIRCAAI / ITI VNU Hanoi.
Une affiliation ne permet pas de déduire la nationalité d’un auteur.