Experiments on Page Rank Algorithm in the XML Information Retrieval Context

Abstract

In this paper we present two adaptations of the PageRank algorithm to collections of XML documents and the experimental results obtained for the Wikipedia collection used at INEX-1 2007. These adaptations to which we referred as ldquoDOCRANK and TOPICAL_docrankrdquo allow the re-rank of the results returned by the base run execution to improve retrieval quality. Our experiments are performed on the results returned by the three best ranked systems in the ldquoFocusedrdquo task of INEX 2007. Evaluations have shown improvements in the quality of retrieval results (improvement of some topics is very significant, eg: topic 491, topic 521, etc.). The best improvement achieved in the results returned by the DALIAN2 university system (global rate obtained for the 107 topics of INEX 2007) was about 3.78%