Datasets:

nielsr

/

funsd-layoutlmv3

like 31

I have an idea, but it would be nice to have documentation to confirm this.

Based on the JSON format of the original dataset (see the example on page 2) and the script used to create this dataset, the prefix B- distinguishes the first word of a label (question, answer, header) and the prefix I- is given to any remaining words for that label. In the example from the paper Registration No. 533, Registration No. is a question and 533 is the answer; in this dataset, they would be labeled as follows (presented as a dictionary because markdown formatting is nice):

{
    "Registration": "B-QUESTION",
    "No.": "I-QUESTION",
    "533": "B-ANSWER"
}

Thanks @tomityrrell got it. Very clear explanation