Five-dimensional context for auditable recommendation

This deposit accompanies the manuscript Five-Dimensional Spatiotemporal–Context Modeling for Auditable Sequential Recommendation: Coupling, Patterns, and Cross-Domain Validation. It contains (1) a snapshot of the Python experiment code used for offline evaluation (SASRec-class backbone, proposed NA-5D extensions, pattern mining, and controls such as H4 permutation), and (2) frozen JSON exports of the reported metrics and diagnostics, so readers can verify tables and appendix numbers without retraining.The programs/experiments/ folder holds entry points (run_prelim.py, complete_missing.py, run_yelp_geo.py, run_sensitivity_sweep.py, run_baselines.py, run_merged_gamma_study.py, run_coarse_branch_ablation.py), supporting modules, config_default.yaml, and requirements.txt. Scripts expect to reside under a project root next to downloaded public datasets (MovieLens, HetRec 2011, Yelp Open Dataset); raw data files are not redistributed here—see the paper’s data availability statement for URLs and licenses.The results/ folder stores machine-readable outputs such as prelim_results.json, complete_results.json, yelp_geo_results.json, sensitivity_grid.json, merged_gamma_study.json, coarse_branch_ablation.json, and baseline_results.json, plus optional auxiliary JSON from smaller dev runs.Environment: Python 3 with PyTorch (CUDA optional but recommended). Use pip install -r experiments/requirements.txt after placing experiments/ in the main repository layout described in programs/README.txt. Keywords: sequential recommendation; context modeling; pattern mining; reproducibility; MovieLens; HetRec; Yelp Open Dataset; PyTorch; offline evaluation