Study on XML Retrieval Results Classification

Abstract

Mining (classify or clustering) retrieval results to serve relevance feedback mechanism of search engine is an important solution to improve effectiveness of retrieval. Unlike plain text documents, since the XML documents are semi-structured data, for XML retrieval results classification, consider exploiting structure features of XML documents, such as tag paths and edges etc. We propose to use Support Vector Machine (SVM) classifier to classify XML retrieval results exploiting both their content and structure features. We implemented the classification method on XML retrieval results based on the IEEE SC corpus. Compared with k-nearest neighbor classification (KNN) on the same dataset in our application, SVM perform better. The experiment results have also shown that the use of structure features, especially tag paths and edges, can improve the classification performance significantly.