In vector representations unknown words are sometimes modeled by having all
low-frequency words map to a string like "unk" during training. Right now
unknown words are handled by always returning 0 similarity. If a map passed in
has an "unk" string then use it when the words passed in are not in the map.
Original issue reported on code.google.com by tim.mil...@gmail.com on 3 Feb 2015 at 10:22
Original issue reported on code.google.com by
tim.mil...@gmail.com
on 3 Feb 2015 at 10:22