hkdse-ict-evaluation

To evaluate AI grading accuracy against ground truth levels, compute rank-based correlation metrics like Spearman ρ and Kendall τ to measure the relationship between predicted levels and true levels. This process identifies exact and adjacent match rates to validate the grading outputs.

Spearman ρ and Kendall τ rank correlation metrics are used to compare predicted grades with ground truth data. These metrics quantify the rank relationship, while exact and adjacent match rates provide additional accuracy validation for the grading evaluation.

Yes, you can generate DOCX reports for individual student grading evaluations using the python-docx dependency. The evaluation workflow outputs per-student DOCX reports alongside aggregate metrics and evaluation figures to support auditing and validation of the grading results.

To validate HKDSE ICT AI grading outputs, compare AI-generated grades against ground-truth mappings for the 2025 dataset. The evaluation processes grading outputs for 10 students, producing per-student and aggregate metrics including Spearman ρ, Kendall τ, and match rates.

Yes, python-docx is required as a dependency to generate DOCX reports for the statistical evaluation of grading. It enables the creation of per-student DOCX reports that document the Spearman ρ, Kendall τ, and match rate metrics computed during the evaluation.

The evaluation supports optional ground-truth and level-division data to adapt to different evaluation setups. If ground-truth mapping data is missing, the evaluation cannot compute Spearman ρ, Kendall τ, or match rates, as these metrics require true levels for comparison.