Database Reference
In-Depth Information
1.2
10 1
10 0
1.0
10 -1
0.8
10 -2
0.6
Filtered clicks
Incremental filtered clicks
Quality score
Incremental quality score
Normalized fraud score
10 -3
0.4
10 -4
0.2
10 -5
10 -6
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
reshold level
FIGURE 14.10
Single bucket analysis.
score, and the quality score with the percentile threshold set by the histogram filter for
this bucket is studied. Moreover, the number of incremental abusive clicks, that is, the
number of abusive clicks detected solely by the IP size histogram filter and not by other
systems, as well as the incremental quality score, that is, the quality score associated
with the incremental abusive clicks, are analyzed. As clear from Figure 14.10, there is a
sweet spot around 0.7 that identifies a small fraction of clicks, about 1% of the total num-
ber of clicks in this buckets, which have both high fraud score and low quality score.
14.5.4.7 Performance over Time
Figure 14.11 shows the IP size histogram filter performance over a month, where the
filter was run daily. The fraud score and quality score of the filtered click traffic were
0.9
0.8
Normalized fraud score
Filtered quality score
Unfiltered quality score
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0 0
5
10
15
20
25
30
Day
FIGURE 14.11
Fraud score and quality score for different days.
Search WWH ::




Custom Search