Impact of Ngrams-based indexing on XML retrieval

Abstract

We present in this paper a statistical approach of term clustering. This approach is based on a statistical analysis of NGrams shared by a pair of terms and is inspired from the t f × idf criterion commonly used in information retrieval. Being statistical, the approach is completely independent from the lexical and grammatical characteristics of the language in which documents to be indexed are written. Classical indexing is often based on stemming, which consists of transforming a term into its radical. This allows to provide large issues for customized information access. As for us, we consider that this can be made by building term clusters and perform information retrieval based on this concept. This approach is used for XML retrieval, therefore some experiments have been undertaken into a dataset provided by INEX to show its impact compared to Porter stemming method.