On Using Social Context to Model Information Retrieval and Collaboration in Scientific Research Community - Community-Built Databases: Research and Development - page 150

Fig. 6.4 Tuning the parameter

a

0,12

0,1

0,08

p@0.1
p@0.2

0,06

0,04

0,02

0

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

α

Evaluating Search Performance

We compare our model with three baselines detailed as follows:

l TF

IDF model : denotes a traditional information retrieval system implemen-
ted by A PACHE L UCENE based on the TF IDF metric and using the stemming

algorithm SnowBall Stemmer . We used this retrieval system with the same
configuration in our model to select documents and compute their topical
relevance.
l PR-Docs model : denotes a retrieval system that estimates the importance of
documents based on their authority. It combines the topical relevance and the
PageRank score of documents computed on the document graph where edges
represent citation links. Final document relevance is computed as follows:

Rel ð d Þ¼ a RSV ð q ; d Þþð 1 a Þ PageRank docs ð d Þ : (6.9)

We note that the topical relevance RSV( q , d ) is computed using the first baseline
TF parameter on the search effectiveness

and we note that best retrieval precisions are obtained with

IDF. We studied the impact of the a

a ¼ 0.7 for p @0.1

0.3 for p @0.2 as shown in Fig. 6.5 .
l Kirsch's model : denotes the social information retrieval model introduced in
[ 51 ] that represents authors using a binary coauthorship network and computes
their social importance score using the PageRank measure. This model com-
bines the topical relevance and the social relevance as follows:

a ¼

and

Rel ð d Þ¼ RSV ð q d Þ r d ; (6.10)

;

with r d is the social relevance of document d computed as the sum of its authors
PageRank scores.

Figure 6.6 compares results obtained by the different baselines and our social
model tuned with

a ¼

0.5 and

a ¼

0.6 noted, respectively, SM 0.5 and SM 0.6.