On the Usage of Structural Information in Constrained Semi-Supervised Clustering of XML Documents