Information Technology Reference
In-Depth Information
the other hand, compound terms in the domain vocabulary may be suggested
as narrower or related terms as explained in the term suggestion with domain
vocabulary in Figure 4(a). To put it another way, compound nouns in the do-
main vocabulary facilitate implicit use of semantic relations during the dynamic
term suggestions, and provides the users with the semantic context entailed in
the school documents in the portal.
3.3 Term Weighting
In the auto-suggest interface, the suggested terms are ranked in the order of
significance. The equation (1) below is a domain-dependent metrics that exploits
domain vocabulary and topic classification scheme as domain knowledge. Weight
of a term
i
(
W
i
) is calculated as arithmetic average of
X
i
and
Y
i
,whichis
multiplied by
Z
i
.
(
X
i
+
Y
i
)
Z
i
W
i
=
(1)
2
Y
i
mea-
sures the infrequency or rarity of a term's occurrence in a subset of documents
that are classified into the same topic. This is based on the assumption that the
less likely a term is to be associated with topics through documents the better
is it likely to be at discriminating relevant from irrelevant a subset of documents
in the same topic. Moreover, the arithmetic average is multiplied by
X
i
here considers if a term is included in the domain vocabulary, and
Z
i
in order
to reduce significance of terms that are likely to be weak in semantics and occur
frequently. Further details of each term in the equation (1) are given as follows.
X
i
=
A
N
/I
N
,
i
if term
is included in domain vocabulary
0
,
otherwise
where
A
N
is the number of terms in domain vocabulary;
I
N
is the number of terms in documents
and
S
N
1
S
N
D
ki
D
i
log
D
ki
D
i
Y
i
=
−
k
=1
where
S
N
is the number of topics;
D
i
is the number of
documents containing term
i
;and
D
ki
is the number of
documents containing term
i
andclassifiedintopic
k
Z
i
=
0
.
5
,
if the number of hiragana characters in term
i
≤
2
1
,
otherwise
4 Evaluation
The aim of this study is to investigate how effectively users can retrieve school
documents with the support of auto-suggest interface as well as the domain-
dependent term-weighting method. The study used a 2x2 mixed factorial de-
sign with one between-participants variable and one within-participants variable.
