Information Technology Reference
In-Depth Information
Fig. 16.9.
The added-value of the context-aware method in approaching the optimal
k
- application on DS1 and DS2 for document clustering
Fig. 16.10.
The added-value of the context-aware method in approaching the optimal
k
- application on DS3 and DS4 for word clustering
optimal solution. The contribution is made clear when observing, for instance,
the performance of the
C
4 index in DS2: While the optimal solution in terms
of
C
4 is found at
k
=57with
F
=0
.
49, using the index as stopping criterion
entails a first drop to occur at
k
= 655 with
F
=0
.
21. However, when adding
context-awareness to the process, a first drop occurs closer at
k
= 302 with
F
=0
.
354. Similarly, consider the
H
3 index in DS3: While the optimal solution
is found at
k
=53with
F
=0
.
21, using the index alone entails a first drop to
occur at
k
= 418 with
F
=0
.
146. However, when adding context-awareness to
the process, a first drop occurred at
k
= 102 with
F
=0
.
197, which is indeed a
promising result.
16.5.2
On the Quality of the Optimal Solutions
Since a context-aware algorithm is no more taking the merging decisions that op-
timizes
VI
, one may imagine that the method, although approaching the optimal
solution, can deteriorate the quality of this solution. However, results show the
opposite. Actually, the histograms illustrating the
FScore
at the optimal value of
