[1990] LSA(Latent Semantic Analysis) Singular Value Decomposition : SVD - real or complex matrix를 factorization - A = $U ∑ V^T$ - A = mxn (m>n) / $U$ = mxm / $V^T$ = nxn Properties of SVD - Singular vectors of the matrix U and V are orthogonal - The number of positive singular values in ∑ = Rank(A) Reduce SVDs - Thin SVD : ∑ → square matrix - Compact SVD : remove zero-singular values and corresp..
