Applications of Latent Semantic Analysis
Thomas K. Landauer
Abstract
Latent Semantic Analysis (LSA) treats language learning and
representation as a problem in mathematical induction. It casts the
passages of a large and representative text corpus as a system of
simultaneous linear equations in which passage meaning equals the sum of
word meanings. Solution by Singular Value Decomposition (SVD) and
dimension reduction produces a high-dimensional vector representing the
average contribution to passage meanings of every word, and thus of the
similarity between any two passages. LSA simulates human language
understanding with surprising fidelity. Combining LSA with other
statistical language modeling methods increases its practical scope. A
variety of tests and applications illustrate its power, limits, and
raise interesting theoretical issues.
Full Paper (PDF)
|
