machine learning - Aspect-based text labeling for named entity recognition in AWS sagemaker

Question

I'd like to know the best way to label text in a machine learning training data set. I'm using Amazon Sagemaker ground Groundtruth to train named entity recognition (NER). Here's an example:

I walked for 30 minutes and then slept for 60 minutes.

In this example, I'd like the NER to recognize that the action "walked" is associated with "30 minutes" and "slept" is for "60 minutes"

If Sagemaker permitted nested labels, then one strategy I could do is to label "walked for 30 minutes" as an event, then nested under that this event label, "walk" could be labeled an action and 30 minutes "duration".

As far I can tell, Sagemaker doesn't allow nested labels (all words can only be associated with a single label).

What is a good labeling strategy for a training dataset that that similar numbers (minutes) are associated with different parts of a sentence or part of the text?

Feel free to let me know how I can make this question more clear. I'm new to creating training datasets.

Thank you!

question from:https://stackoverflow.com/questions/65910179/aspect-based-text-labeling-for-named-entity-recognition-in-aws-sagemaker

Answer 1

Waitting for answers