Databases Reference
In-Depth Information
granularity than entire emails, where the thread would be a simple chain from
E
6
to
E
5
,
E
6
to
E
4
and so on.
Answer A
12
Question Q
1
Answer A
11
Answer A
22
(a)
Question Q
2
Answer A
21
Answer A
12
Question Q
1
Answer A
11
(a)
Answer A
22
Question Q
2
Answer A
21
Figure 3.15:
(a) FQG for the dialogue act labeling example in Figure
3.10
. (b) FQG for the same
example when an edge is created only between to fragments if one is below the other in an email.
As mentioned, the FQG is only an approximation of the reply relations between fragments.
In some cases, proximity may not indicate any connection and in other cases a connection can exist
between fragments that are never adjacent in any email in the thread. Nonetheless,
Carenini et al.
[
2008
] showed that considering the FQG can be beneficial in email summarization and we argue
that similar benefits could be obtained if the FQG was used to support the other text mining tasks
discussed in this chapter. For instance, let us go back to the dialogue act modeling example shown
in Figure
3.10
. The FQG for that example would be the one shown in Figure
3.15
(a). Remarkably,
if we had identified the fragment asking the questions, it would be straightforward to identify most
of the corresponding answers on the FQG. The only ambiguous case would be
A
11
, which could be
an answer to both
Q
1
and
Q
2
. One heuristic to resolve this problem could be to build the FQG in
a more conservative way, namely, by creating edges in the FQG between to fragments only if one is
below the other in the text, which would result in the FQG in
3.15
(b).
Search WWH ::
Custom Search