Two-Stage Grading
Transparent by design โ every point is traceable to a specific text comparison.
Stage 1: Rule-Based Floor
Calculates a guaranteed minimum score based on literal technical term coverage. Students cannot score below what they explicitly mentioned.
Stage 2: Semantic Grading
Applies the PMC12171532 equations using NLP metrics and Semantic Similarity to award points for meaning and conceptual understanding.
Scoring Metrics & Weights
Semantic Similarity (50%)
TF-Cosine proxy for USE contextual meaning.
Jaccard Similarity (15%)
Token-set intersection for vocabulary usage.
Cosine Frequency (15%)
Word frequency vector angle for synonymous phrasing.
Normalized Word Count (15%)
Ratio-based length checks to prevent inflation.
Edit Similarity (5%)
Normalized Levenshtein for typo tolerance.
๐ Equations from PMC12171532
// Confidence Score (C)
C = 0.5 * Stf + 0.5 * (0.15*Sj + 0.05*Se + 0.15*Sc + 0.15*Sw)// Final Scoring Rule (F)
if (Stf < 0.2) return 0.0if (Stf >= 0.9 && Sw >= 0.8) return 1.0
else return C