Information Technology Reference
In-Depth Information
Table 3. Parameters and their default values
PARAMETER
DEFAULT VALUE
DESCRIPTION
Probability Threshold
0.006
Minimum probability of a shortcut
Depth Bound
5
Maximum length of suffix
Number of Shortcuts
3
Number of top shortcuts recommended
N
2
Order of N-gram model
testing set for MINCOST, while the remaining portions are used as training set for building the N-gram
model as described in the third section. The results are averaged over these ten runs.
The parameters and their default values are given in Table 3.
The user behavior when provided with shortcuts is simulated by making two assumptions. First, it
is assumed that when presented with one or more shortcuts that lead to destinations along the user's
session, the user will select the shortcut that leads farthest along the session. Second, when no shortcuts
lead to pages in the user's session, the user will follow the next link in the session.
experimental r esults from nASA dataset
The results from experiments on NASA data set are presented in this section. Similar results are obtained
from experiments on EPA data set, and are thus omitted.
Figures 2 and 3 show the average cost saved and percentage of average cost saved with respect to
probability threshold, respectively. It is observed that both measures increase with decrease in probabil-
ity threshold and stabilize around 0.0080. This is because there are not many shortcuts with significant
savings after this value.
From Figures 2 and 3, it is obvious that the two performance measures, average cost saved and per-
centage of average cost saved, react similarly to changes in probability threshold. It is not a coincidence.
Other experiments also reveal the high correlation of these two measures. In the rest of this section,
only the results for percentage of average cost saved are presented.
As shown in Figure 4, the percentage of average cost saved is not affected by depth bound except
when it increases from 1 to 2. Just like MINPATH, MINCOST is more sensitive to probability threshold
than depth bound. Depth bound has a larger impact when the probability threshold is large. However,
for most reasonable probability thresholds, depth bound's effect is overshadowed by that of probability
threshold.
As expected, the percentage of average cost saved increases with the number of shortcuts recom-
mended by MINCOST, as shown in Figure 5. The increase in percentage of average cost saved is espe-
cially significant when the number of shortcuts increases from 1 to 2 and 2 to 3. However, the increase
is much smaller for larger numbers of shortcuts, for example, when the number of shortcuts increases
Search WWH ::
Custom Search