More than 5000 words in 'train.tsv' can't be found in 'glove.840B.300d.txt'

[furukawayuan-Yao]

When I omited the overlap part in dev and test datasets, I found that more than 5000 words in training dataset couldn't be embedded by Glove. These words obtained the zero-vector embedding . I inspected and found that there are many words mispelled like 'neext' , 'somethinghe', ect. What should we do to process such data?Can the negative effect of these wrong-spelled words be offset by the option of 'word_freq'? Looking forward to your reply. Thank you very much😀！