This code doesn't work well by using English word2vec predefined vectors

[sasikumar-enugunta]

I've just used the same approach, replacing GoogleNews-vectors-negative300.bin instead of Indonesian word2vec.
I've created the train, val and test.tsv files similar to the data which used for this code.
The dataset which I have is a combination of 85% English and 15% Indonesian words.
But the code is predicting everything as 'O' tag.
Could you please help me out to resolve this issue?