Question : Well, with dense vectors such as Continuous Bag of Words, Skip-gram model, what parameter  value should be used for parameters such as number of dimensions, window size e.t.c?
Answer: The parameter  value selection depends on the  task at hand, i.e whether you want to capture semantic information or syntactic information.
Syntactic information such as live – lives -lived.
And semantic information as “Japanese restaurant” and “Sushi  Shop”.

CBOW skip-gram parameter selection.JPG

From the above figure, it is clear that the best parameter settings is,
Dimension ~ 300
Window Size : 8

Advertisements