Dimension-reduced estimation of word co-occurrence probability
Kilyoun Kim, Key‐Sun Choi · 2000
We investigate a novel approach to solve the problem of sparse data through dimension reduction. Linear algebraic technique called LSA/SVD is used to find co-relationships of sparse words. Three variant estimation methods are suggested and they are evaluated for estimating unseen noun-verb co-occurrence probability. The model shows possibility to be alternative probability smoothing method.