#truncated-svd

Question 1: Select all that apply

Which statements about latent semantic analysis (LSA, i.e. truncated SVD), PCA and non-negative matrix factorization (NMF) are true? Select all that apply.

Unlike a bag-of-words vector, LSA gives a short and dense representation of a document1
In LSA, the weight vectors in $W$ can be read as topics extracted from the documents2
In LSA, a document's row of the transformed data $Z$ says how strongly each topic is present in it3
On a 10-dimensional dataset, PCA with 4 components and PCA with 1 component give the same first component (up to sign)4
On a 10-dimensional dataset, NMF with 4 components and NMF with 1 component give the same first component5

Timed test