Another query:
Example of precision and recall
Based on the above queries and pooling.
Lycos | ||||||||
q1 | q2 | q1 | q2 | |||||
P | R | P | R | P | R | P | R | |
Top 1 | - | - | 1/1=1 | 1/8=0.13 | 1/1=1 | 1/7=0.14 | 1/1=1 | 1/8=0.13 |
Top 2 | 1/2=0.5 | 1/7=0.14 | 2/2=1 | 2/8=0.25 | 1/2=0.5 | 1/7=0.14 | 2/2=1 | 2/8=0.25 |
Top 3 | 1/3=0.33 | 1/7=0.14 | 2/3=0.67 | 2/8=0.25 | 2/3=0.67 | 2/7=0.29 | 3/3=1 | 3/8=0.38 |
Top 4 | 2/4=0.5 | 2/7=0.29 | 3/4=0.75 | 3/8=0.38 | 3/4=0.75 | 3/7=0.43 | 3/4=0.75 | 3/8=0.38 |
Top 5 | 3/5=0.6 | 3/7=0.43 | 4/5=0.8 | 4/8=0.5 | 3/5=0.6 | 3/7=0.43 | 3/5=0.6 | 3/8=0.38 |
Top 6 | 4/6=0.67 | 4/7=0.57 | 5/6=0.83 | 5/8=0.63 | 3/6=0.5 | 3/7=0.43 | 3/6=0.5 | 3/8=0.38 |
Top 7 | 5/7=0.71 | 5/7=0.71 | 6/7=0.86 | 6/8=0.75 | 3/7=0.43 | 3/7=0.43 | 3/7=0.43 | 3/8=0.38 |
Top 8 | 6/8=0.75 | 6/7=0.86 | 6/8=0.75 | 6/8=0.75 | 3/8=0.38 | 3/7=0.43 | 4/8=0.5 | 4/8=0.5 |
Top 9 | 6/9=0.67 | 6/7=0.86 | 6/9=0.67 | 6/8=0.75 | 3/9=0.33 | 3/7=0.43 | 4/9=0.44 | 4/8=0.5 |
Top 10 | 6/10=0.6 | 6/7=0.86 | 7/10=0.7 | 7/8=0.88 | 3/10=0.3 | 3/7=0.43 | 4/10=0.4 | 4/8=0.5 |
Please observe that recall values are always rising as we consider top 1, top 2, etc. documents. Precision values can go up and down. Also observe that the top 1 precision-recall values for q1 on Google are not defined because the top 1 link was not relevant.
Please observe that in Assignment 1 you measure precision-recall for the top 5, top 10, top 15 ... top 30 links.
Example of interpolated precision
Based on the above precision-recall values:
Google, interpolated precision | Lycos, interpolated precision | |||||
Standard recall | q1 | q2 | Average | q1 | q2 | Average |
0.1 | 0.75 | 1 | 0.88 | 1 | 1 | 1 |
0.2 | 0.75 | 1 | 0.88 | 0.75 | 1 | 0.88 |
0.3 | 0.75 | 0.86 | 0.81 | 0.75 | 1 | 0.88 |
0.4 | 0.75 | 0.86 | 0.81 | 0.75 | 0.50 | 0.63 |
0.5 | 0.75 | 0.86 | 0.81 | 0 | 0.5 | 0.25 |
0.6 | 0.75 | 0.86 | 0.81 | - | - | - |
0.7 | 0.75 | 0.86 | 0.81 | - | - | - |
0.8 | 0.75 | 0.7 | 0.73 | - | - | - |
In order to enable average, missing interpolated precision values are replaced with 0.
Please observe that an interpolated precision value is the highest, not closest, measured precision value "to the right" (which includes your exact measured precision if your measured recall value happens to be equal to the standard recall value being considered). A sequence of interpolated precision values is either flat or falling, never rising.
Each curve represents the average interpolated precision-recall curve, one curve for Google, one curve for Lycos.