Every day more digital data in semi-structured format are available on the World Wide Web, corporate intranets, and other media. Knowledge management using information search and processing is essential in the field of academic writing. This task becomes increasingly complex and defiant, mainly because collections of documents are usually heterogeneous, big, diverse, and dynamic. To resolve these challenges it is essential to improve management of time necessary to process scientific information. In this paper, we propose a new method of automatic clustering of XML documents based on their content and structure, as well as on a new similarity function OverallSimSUX which facilitates capturing the degree of similarity among documents. Evaluation of our proposal by means of experiments with data sets showed better results than those in previous work.