Testing the polyrepresentation principle by performing fusion of INEX results

Abstract

This thesis seeks to test the theory of polyrepresentation by performing data fusion. The theory provides a framework for investigating overlap between different representations. Further, it hypothesises that overlap between these contain a higher number of relevant documents than one representation alone. In this case, the representations are different result sets which are combined in order to examine the new results in comparison with each original result set. Four different result sets for various XML retrieval approaches were selected from INEX. In evaluation, performance measures such as normalised cumulated gain, effort-precision/gain-recall, and average values for these measures were calculated. Statistical tests were carried out on the mean average values for the results. The results show that improved performance is achieved, especially for one of the approaches used in INEX for element retrieval.