Figure 1. Singular Value Decomposition (SVD) on a term document matrix and the generation of the U* matrix.

(A–C) U and V^T contain the LSI vectors for terms and documents, respectively while Σ contains the singular values of the original term document matrix. (D) An illustration of the resulting matrix U*, obtained by the multiplication of U_k and P. Note that the resulting matrix contains the word vectors and phrase vectors in LSI space, facilitating the comparison between every word/phrase and every other word/phrase entity.