Database Reference
In-Depth Information
Table 4.2
An example for calculating average Qrecall and average hit-rate.
t
[
k
]
Place in list (j)
Positive probability
Qrecall
Hit-rate
1
0.45
1
0
.
25
1
2
0.34
0
0
.
25
0
.
5
0
.
5
0
.
667
3
0.32
1
0
.
75
0
.
75
4
0.26
1
5
0.15
0
0
.
75
0
.
6
6
0.14
0
0
.
75
0
.
5
7
0.09
1
1
0
.
571
8
0.07
0
1
0
.
5
9
0.06
0
1
0
.
444
10
0.03
0
1
0
.
4
Average:
0
.
893
0
.
747
Table 4.3
Qrecall and Hit-rate in an optimum prediction.
t
[
k
]
Place in list (j)
Positive probability
Qrecall
Hit-rate
1
0.45
1
0
.
25
1
2
0.34
1
0
.
5
1
3
0.32
1
0
.
75
1
4
0.26
1
1
1
5
0.15
0
1
0
.
8
6
0.14
0
1
0
.
667
7
0.09
0
1
0
.
571
8
0.07
0
1
0
.
5
9
0.06
0
1
0
.
444
10
0.03
0
1
0
.
4
Average:
1
1
the head of the list. This case is illustrated in Table 4.3. A summary of the
key differences are provided in Table 4.4.
4.2.6.9
Potential Extract Measure
(
PEM
)
To better understand the behavior of Qrecall curves, consider the cases of
random prediction and optimum prediction.
Suppose no learning process was applied on the data and the list
produced as a prediction would be the test set in its original (random)
order. On the assumption that positive instances are distributed uniformly
in the population, then a quota of random size contains a number of
positive instances that are proportional to the
apriori
proportion of positive
instances in the population. Thus, a Qrecall curve that describes a uniform
distribution (which can be considered as a model that predicts as well as a
Search WWH ::
Custom Search