1. Confusion Matrix(混淆矩阵)
Relevant | Nonrelevant | |
---|---|---|
Retrieved | True Positive | False Positive |
Not Retrieved | False Negative | True Negative |
2. Precision and Recall
Precision = TP/(TP+FP),即被检索出的文档中真正相关的文档所占的比例。
Recall = TP/(TP+FN),即真正相关的文档中被成功召回(检索)的文档所占的比例。
下图解释较为形象:
3. F measure
We normally use a balanced F1 measure with β=1.
4. Rank-Based Measures
Slides: Stanford CS276 Ranked-Based Measures
- Precision@K
- Mean Average Precision (MAP)
- Mean Reciprocal Rank (MRR)
- Normalized Discounted Cumulative Gain (NDCG)