Small-worlds clustering applied to documents re-ranking

We propose in this paper an approach for document clustering. It consists on representing the corpus as a document graph, where links are defined by some criteria. These links are quantified by similarity measures. We aim to join this context with clustering to build small-world networks of homogeneous documents. The homogeneity of the clusters is measured according to the properties of small-worlds networks. The clusters, as well as their properties, allow to re-rank search results. Some experiments have been undertaken into a corpus provided by TREC and the obtained results show the contribution of small-worlds properties and analysis in information retrieval and document re-ranking.