Database Reference
In-Depth Information
authors in the social network of bibliographic resources. Therefore, we compute for
each author a social importance score
C
G
(
a
i
) using one of the following importance
measures: the
Betweeness
, the
Closeness
, the
PageRank
, the
HITS
's
Authority
score, and the
HITS
's
Hub
score. We apply these importance measures only on
the subgraph of authors
G
a
¼
A
). Edges in
G
a
denote
either coauthorships or either citation links and weighted as described previously.
Afterward, a social importance score is transposed from authors to documents
using a weighted sum aggregation as follows:
(
A
,
E
a
), where
E
a
(
A
X
k
Imp
G
ð
d
Þ¼
w
ð
a
i
;
d
Þ
C
G
ð
a
i
Þ
:
(6.7)
i¼
1
The social score of document Imp
G
(
d
) estimates its social relevance. Neverthe-
less, it is not enough to retrieve relevant documents from the collection. Therefore,
we combine Imp
G
(
d
) score with a traditional information retrieval metric such a
TF-IDF score as follows:
Rel
ð
d
Þ¼
a
RSV
ð
q
d
Þþð
1
a
Þ
Imp
G
ð
d
Þ
;
(6.8)
;
where
1] is a weighting parameter, RSV(
q
,
d
) is a normalized similarity
measure between query
q
and document
d
, and Imp
G
(
d
) is the importance of
document
d
in the social network
G
.
a
2
[0
...
6.5.3 Experimental Evaluation
In order to evaluate the effectiveness of our model, we conduct a series of experi-
ments on a scientific documents dataset published for the ACM SIGIR conference
from 1978 to 2008. The main evaluation objectives are to:
1. Compare different importance measures with both binary and weighted social
network models to estimate importance of scientific papers.
2. Evaluate the effectiveness of our model compared with traditional information
retrieval models and other closely related retrieval models.
6.5.3.1 Experimental Datasets and Design
We used for our experiments the SIGIR dataset that contains information about
authors and citation links in addition to the textual content of publications. We
included in the social network all authors having published at least one paper for the
ACM SIGIR conference. Two authors are associated with a social relationship if
they either coauthored a SIGIR publication or one of them cites the other author
through his SIGIR paper.
Search WWH ::
Custom Search