How to Build and Deploy a Custom NER Model with CoNLL Data Using Azure ML Designer?

Question

I am working on a Custom Named Entity Recognition (NER) project using Azure ML Designer. I have already labeled resume text files in CoNLL format (e.g.,:

Curriculum O
Vitae O
veera B-Candidate_name
nandan I-Candidate_name
Contact O
No O
- O
123456789 B-Mobile_no
Email O
- O
******@gmail.com B-Email_id
PROJECTS O
Working O
as O
Software B-Designation
Engineer I-Designation
in O
abc B-Companies_Worked_At
Tech I-Companies_Worked_At
Limited I-Companies_Worked_At

) and exported them as labelled.CoNLL.txt.

I want to build a pipeline using Azure ML Designer to train and deploy a custom NER model with this labeled dataset.

My Questions:

1. Which modules/components should I drag and drop in Azure ML Designer to preprocess, train, and deploy the NER model using this CoNLL dataset?
2. How should I configure the connections between these modules to achieve a working pipeline?
3. Is a GPU compute cluster required for training, or is a CPU cluster sufficient for this NER model in Azure ML Designer?
4. After training, what components should I use to create and deploy the inference pipeline?

Looking forward to detailed guidance from the community. Thanks!

Answer 1

Hi Manas mohanty, I don't see anywhere in the azure machine learning about custom ner with own labelled data for text files. I am not looking out to train and deploy the model using automl. I want to do it using designer interface.so once exported conll.txt file using this file and using CPU instances how we can train and deploy a custom ner pipeline.

Answer 2

Hi Dinnemidi Ananda Kumar,

Sorry for the delay in response.

You can create a custom NER model using Language studio and consume the endpoint.

We need to have NER model either from AutoML or Tensorflow/Pytorch to tune the model with custom dataset to use in Design studio.

Below is the process to do it in Designer studio.

1. Use data component to import data asset, configure the UI to connect to storage points.
2. Use model component to connect to with existing model or custom model component
3. You can import the training pipeline from Studio and create inference pipeline out of it.

Reference on using UI designer

Tutorial on Auto price prediction with Designer

Deploy for inference from Designer Studio

Thank you.