Significant term extraction by higher order SVD

In this paper, we present a novel method for term importance, called Tensor Term Indexing (TTI). This extracts significant terms from a document as well as a coherent collection of document set. The basic idea of this approach is to represent the whole document collection in a Term-Sentence-Document tensor and employs higher-order singular value decomposition (HOSVD) for important term extraction. TTI uses the lower rank approximation technique to reduce noise by eliminating anecdotal terms, to mitigate synonymy by merging the dimensions associated with terms that have similar meanings, and to mitigates polysemy, since components of polysemous words that point in the "right" direction are added to the components of words that share a similar meaning. Our evaluation shows that that TTI model can extract significant terms relevant to a topic from a small number of documents which Term Frequency and Inverse Document Frequency (tfidf) cannot.