tstanislawek / awesome-document-understanding Goto Github PK

Hello,

I know this isnt a issue, but i couldnt find a better place to ask this question.

I guess extracting data from resumes belongs to the key-information-extraction area.
So for the start I thought about using just a normal BERT and in my training data I only mark the entities that want to extract. But does it also makes sense to create a label for the label "english" (see example below) to get better results or to use relation extraction at this semi-form like data? Or does the behaviour of BERT recognizes automatically that after the string "english: " is going be a grade?

For a simple example at extracting grades from a resume:
input:
"english: 2"
-> do i need only need to annotate "2" or is it recommended to do something else

output:
grade_english: "2"

wanted labels:

• firstname
• lastname
• last_job_title
• graduation
• grade_english
• grade_math
• grade_economy

I am sorry, I know this is not an issue, but I don't know where to ask it.

I am parsing PDF documents and now I have a task to group entities together: I have a chemical and its characteristics, I am parsing them using NER (huggingface transformers) and quality is OK, but I don't know how to group each chemical with corresponding characteristics (I don't even how the task is called). I can write some rules, that characteristics, which appear after the chemical name, correspond to this chemical, but sometimes the order is different and some characteristics appear before the name of the chemical.

So I want to use some model to link chemicals and their corresponding characteristics somehow together.

Please can you help me and give me some advice for this problem