Abstract: Due to its semi-structured characteristic XML document implies rich term semantics in the structure information. It is inappropriate to measure XML term weight by general Term Frequency-Inverse Document Frequency (TF-IDF) approach. This is because that, when measuring the XML term weight, this approach only considers the term frequency and document frequency but ignores the term semantics implied in the structure information. A novel term weight measuring approach is proposed to overcome the above shortcoming and improve the performance. This new approach takes the factors affecting term semantics into account, such as the paths which contain terms, term frequency in a certain path, frequency of document which contains a certain path, depth of path, etc. Experimental results on several datasets show that, compared with TF-IDF and rule based approaches, the proposed approach can improve the recall, the precision and F₁-measure in the classification of XML documents.

Key words: computer software, semi-structured data, XML mining, XML classification, term semantics, weight measurement