Volume : V, Issue : IV, April - 2015

On Score Normalization in Distributed Information Retrieval

Benjamin Ghansah, Benuwa Ben Bright

Abstract :

 This paper presents a novel result merging technique that uses the idea of score normalization for curve 

fitting. First the algorithm computes the cumulative score distribution from an ideal distribution: we used 
the central sample database for this purpose. Second we used the exponential and Gaussian distributions to build 
an optimal score distribution (OSD). Third we normalize the scores returned by individual information source by mapping each score to the OSD. Lastly the normalized values obtained are used as global scores for result merging. We 
compare our results with SAFE merging algorithm, which showed superior performance on two testbeds; TREC123, 
TREC4kmeans.

Keywords :

Article: Download PDF   DOI : 10.36106/ijar  

Cite This Article:

Benjamin Ghansah, Shengli Wu, Benuwa Ben-Bright On Score Normalization in Distributed Information Retrieval Indian Journal of Applied Research, Vol.5, Issue : 4 April 2015


Number of Downloads : 912


References :