emnlp2017-bilstm-cnn-crf icon indicating copy to clipboard operation
emnlp2017-bilstm-cnn-crf copied to clipboard

Question about tag scheme

Open PengLU1101 opened this issue 6 years ago • 2 comments

Hi UKP team,

I want to try different tag scheme in my work. Could you tell me if there is any difference between IOBES and BIOES in your implementation?

Thanks,

Peng

PengLU1101 avatar Sep 26 '18 13:09 PengLU1101

Hi @PengLU1101 , you can find an explanation of the different encodings in this paper: https://arxiv.org/abs/1707.06799

The network supports BIO, IOB, and IOBES encoding. The names of these encodings are not really fixed in literature. Our BIO encoding is sometimes called IOB and in other cases it is called IOB2. Quite confusing.

IOBES and BIOES is usually the same encoding.

For our implementation I recommend the BIO encoding: B- notes the start of a tag, I- are the following tokens of the same tag and O is used for every token that has no tag.

nreimers avatar Sep 26 '18 15:09 nreimers

Hi @PengLU1101 , you can find an explanation of the different encodings in this paper: https://arxiv.org/abs/1707.06799

The network supports BIO, IOB, and IOBES encoding. The names of these encodings are not really fixed in literature. Our BIO encoding is sometimes called IOB and in other cases it is called IOB2. Quite confusing.

IOBES and BIOES is usually the same encoding.

For our implementation I recommend the BIO encoding: B- notes the start of a tag, I- are the following tokens of the same tag and O is used for every token that has no tag.

Thanks so much.

PengLU1101 avatar Sep 26 '18 16:09 PengLU1101