A combination of Wilcoxon test and R-estimates for document organization and retrieval

The Wilcoxon signed-rank test is exploited for document organization and retrieval in this paper. A novel modeling method for documents and a distance metric between documents are proposed. Both document modeling and document comparisons are based on signed-ranks and are applied to the frequency of occurrence of the document bigrams. A metric using the Wilcoxon signed-rank test exploits these signedranks to assess the contextual similarity between the documents. A variant of the self-organizing map algorithm (SOM), the so-called Wilcoxon SOM, is built. The quality of the document clustering produced by the Wilcoxon SOM against the standard SOM is compared in terms of their average recallprecision rates for document retrieval based on the document clusters formed. The reported improvements in recall and precision are shown to be statistically significant.