Semantic analysis of text: demo of gensim over arXiv.org

aura.fi.muni.cz

10 pointsRadim4 comments

Does they correctly use word "Semantic". It may be OK for algorithm title(Latent Semantic Analysis), but results not so "Semantic", it is usual topic detection based on vector-space model. Usually "Semantic analysis" corresponds to next to syntactic level of analysis, some kind of understanding. We have no such analysis here. Also imho(not only my), word "Semantic" is marker for oil snake trader. Is there some quality measures? This algorithm gives 20% of other topic for random article in average.

You're right, this is "semantics" of the statistical type.

There is no "understanding" involved. Just unsupervised number crunching.

Sorry, but there is no "semantics" of statistical type. You have semantics, or have not.

Fair enough, I see you take the linguist's approach :-)

You might try googling for "statistical semantics" though ( http://en.wikipedia.org/wiki/Statistical_semantics ).

More relevant to HN, there are entire businesses successfully using this (admittedly primitive) notion of "statistical semantics": Chomp.com, SketchEngine.co.uk, SumoBrain.com , PearsonKT.com etc.

Reply on news.ycombinator.com