Source documents in multi-document summarization tasks could not provide order information
directly. But they definitely contain the clue of order, because each of them describes some aspects
of the same topic. There are some reasons to assure that the sequence of sentences in source
documents could be the reference standards of sentence ordering.
To learn the information of sentences sequence in source documents and predict the order of
sentences in summary, we treat it as a classification task. Firstly, we train the model of classification
with the position information of representative sentences in source documents. Secondly, we
predict the sentence position in summary. Support vector machine (SVM) is a kind of supervised
learning method for classification, and we use libsvm as classification tool in our model (Chang
and Lin, 2001).