Examining the impact of stemming on clustering Turkish texts

Preprocessing is an important step in information retrieval and text mining. In this study, we examined the impact of stemming on clustering Turkish texts. We used two datasets compiled from web sites of Turkish news agencies, and performed extensive experiments. We empirically show that there is no significant evidence that stemming always improves the quality of clustering for texts in Turkish. However, when stemming is used, dimensionality of the document-term matrix dramatically decreases without inversely affecting the clustering performance. As a result, it is highly recommended to apply stemming for clustering Turkish texts.