Skip to content

LSA.md

Latest commit

 

History

History
18 lines (9 loc) · 996 Bytes

LSA.md

File metadata and controls

18 lines (9 loc) · 996 Bytes

LSA (Latent Semantic Analysis)

wikipedia, datacamp.com

  • LSA assumes words close in meaning will occur in similar peices of text with distributional hypothesis.

  • LSA tries finding low-rank approximtion for term-document matrix by dot product between all term-vectors (giving correlation between terms) and document-vectors (giving document correlation over terms).

  • From theory of Linear Algebra, there exists a decomposition of term-document matrix giving a diagonal matrix alongwith a tall and a wide matrix. This is SVD.

  • Uses a BoW (Bag of Word) model resulting in term-document matrix (rows for words, columns for documents).

  • LSA learns latent topics by performing matrix decomposition on document-term matrix using SVD.

  • LSA typically used as dimesnion reduction or noise reduction technique.