Latent semantic analysis (LSA) is a statistical technique from natural language processing. It uses how often words occur together across many documents to find hidden relationships between terms and topics. In document retrieval the same method is called latent semantic indexing (LSI).
How does LSA work?
LSA starts with a matrix that records how often each word occurs in each document. A singular value decomposition (SVD) shrinks this matrix to a small number of dimensions. In the resulting space, documents and words can be compared with cosine similarity. A document about “automobiles” can therefore match a query about “cars” even if the word “cars” never appears in it.
Known limits are the computing cost, the choice of the number of dimensions, words with several meanings that are only partly handled, and the fact that word order is ignored.
Where does it come from?
The method was developed by researchers at Bellcore, among them Scott Deerwester, Susan Dumais and Thomas Landauer. US patent 4,839,853 on the technique was granted in 1989 and has expired.
What does LSA mean for SEO?
In SEO, LSA mostly shows up as the idea of “LSI keywords”, meaning related words that supposedly help a page rank. No Google documentation found for this entry says that Search uses LSA to rank pages.
Sources
Last updated: