hkdse-chinese-evaluation

To evaluate HKDSE Chinese AI grading, compare AI predicted scores against ground-truth levels 1–5. The workflow computes Spearman ρ, Kendall τ, exact match rates, and adjacent match rates to quantify grading accuracy.

HKDSE Chinese grading evaluation uses rank-based metrics including Spearman ρ and Kendall τ. It also calculates exact match rates and adjacent match rates to measure alignment between AI predictions and ground-truth levels.

Generate DOCX evaluation reports by running the workflow with the YEAR variable set to the target year. The system processes grading outputs and ground-truth mappings to produce per-student and class-level reports with visualization figures.

You can evaluate a different year of HKDSE Chinese grading by setting the YEAR variable to your target year. The workflow requires corresponding AI grading outputs and ground-truth level mappings for that specific year.

You need AI grading prediction outputs and ground-truth level mappings ranging from 1 to 5. The evaluation workflow aligns these datasets to calculate rank-based metrics and generate per-student reports for 10 students.