Unsupervised zoning of scientific articles using huffman trees

In this report we propose a new method of unsupervised zoning based on Huffman coding trees. The suggested method acts on the level of sentences and obtains a Huffman tree whose upper part is equal to the tree created by the method of argumentative zoning. The proposed method gives a general framework for the unsupervised zoning, and may be straightforwardly transformed to supervised zoning by mapping the bits defined by human annotator into features.