hkdse-subject-evaluation

To evaluate AI grading accuracy for HKDSE subjects, this Skill compares predicted levels against ground-truth labels using ordinal metrics like Spearman ρ and Kendall τ. It computes exact and adjacent match rates, generates visualizations, and outputs DOCX evaluation reports.

Ordinal grade evaluation uses rank-based statistical metrics including Spearman ρ and Kendall τ to compare predicted grades with ground-truth levels. These metrics assess the correlation and rank order of AI-generated HKDSE subject grades.

To generate an evaluation report for AI-assigned HKDSE grades, provide year-specific data and optional rubric level divisions. The workflow then computes accuracy metrics, produces visualization figures, and exports a comprehensive DOCX report.

Yes, you can use optional rubric level divisions to validate AI grading results. The Skill aligns ground-truth data with per-student results and final scores to compute exact and adjacent match rates across the specified rubric levels.

The evaluation workflow supports multiple years of HKDSE grading data by applying year-specific data inputs. It processes subject-graded outputs across the specified year scope to generate comprehensive year-based evaluation reports.

Rank-based metrics for subject grade evaluation may not fully capture the magnitude of grading errors, only rank order. Evaluation is limited by the availability of accurate ground-truth labels and requires year-specific data for precise ordinal comparison.