LSI typically uses both a local and global weighting scheme to increase or decrease the relative importance of terms within documents and across the entire document collection, respectively. The product of the local and global weighting functions is applied to each non-zero element of A,
where 
is the local weighting
function for term i in document j and 
is the global
weighting function for term i. Among the popular local and global
weighting functions tried with LSI
(see [8], p. 233),
Dumais found the log-entropy weighting scheme provided a 
advantage over raw term frequency on several standard document test
collections [8].