Are there any easy ways to convert a small block of text (10-100 words) into a <1000 dimensional feature vector? One solution that I'm considering is doing an n-gram encoding, then using random projections to reduce the dimensionality.
[link][5 comments]