Behaviour-Driven Transition Graph Construction Pipeline from EdNet-KT3 (K3_clean)

This repository provides the full reproducible pipeline and cleaned dataset used to construct behaviour-driven transition graphs from the EdNet-KT3 dataset. Behaviour-driven transitions capture how learners navigate between learning objects based on observed interaction sequences, reflecting actual platform usage rather than predefined instructional structures. The pipeline performs:- Canonical identifier mapping- Burst duplicate removal- Respond-action filtering- Chronological trajectory reconstruction- First-order transition extraction- Frequency-based filtering across thresholds τ ∈ {1, 3, 5, 10, 20} The main results correspond to τ = 3, producing a directed transition graph with:- 11,493 nodes (learning objects)- 1,203,210 edges (stable transitions)- Graph density: 9.11 × 10⁻³- Mean out-degree: 104.69 (95% CI: [101.31, 107.86]) Statistical validation includes:- Kolmogorov–Smirnov test against an Erdős–Rényi baseline (D = 0.658, p < 1 × 10⁻¹⁶), confirming heavy-tailed degree structure- Bootstrap confidence intervals (B = 1,000 resamples)- Sensitivity analysis across multiple τ thresholds The repository includes:- Cleaned dataset (K3_clean)- Full graph construction pipeline- Precomputed outputs (statistics, distributions, sensitivity analysis)- Reproducibility documentation All results reported in the associated manuscript are fully reproducible using the provided resources. --- Version 3.0 - Updated graph construction results (τ = 3)- Fixed node and edge counts (11,493 nodes, 1,203,210 edges)- Added statistical validation (KS test, bootstrap CI)- Added reproducibility outputs and sensitivity analysis