Co-clustering in text mining has been proposed to partition words and documents simultaneously. Although the main advantage of this approach may improve interpretation of clusters on the data, there are still few proposals on these methods; while one-way partition is even now widely utilized for information retrieval. In contrast to structured information, textual data suffer of high dimensionality and sparse matrices, so it is strictly necessary to pre-process texts for applying clustering techniques. In this paper, we propose a new procedure to reduce high dimensionality of corpora and to remove the noises from the unstructured data. We test two different processes to treat data applying two co-clustering algorithms; based on the results we present the procedure that provides the best interpretation of the data.