NLP with a custom model

0

A wants to use a 120-question cognitive assessment program in order to define and describe someone’s thinking preferences, according to 4 quadrants (for example analytic, emotional, creative, organised).

They are interested in adding an NLP feature to their assessment and are thus interested in finding a platform that could provide a classifier model. The aim of this classifier would be to: Diagnose if a piece of free text (answer to an open-ended question) demonstrates high or low preference for a particular quadrant. Ultimately this classifier should be able to analyse communication (emails, documents, job descriptions) Identify the words/strings/characteristics in the text that reflect each quadrant. Be able to use the classifier as en e-mail plugin that would highlight the different words/expression according to the quadrant they represent and suggest recommendations according to the thinking preferences of the recipient of the email.

I was wondering if Amazon Comprehend will meet this requirements for the classifier model because, from what I could see in the FAQs of the service, Comprehend does not support custom models. Could anyone confirm if Comprehend will be a good fit for this use case and, in case it is not, what could be workaround to meet the customer's requirements in the AWS cloud?

asked 6 years ago215 views
1 Answer
0
Accepted Answer

I think Sagemaker seq2seq could provide such a service:

  • If their target is one single label (primary quadrant), they can experiment training seq2seq to predict primary quadrant based on input text sequence
  • If their target is a sequence (score in each quadrant) they could experiment training seq2seq to predict quadrant scores sequence (eg 8/10 creative, 10/10 emotional, etc)

Also, on the security aspect note that non-public free-form text from individuals and emails are usually regarded as very sensitive data; depending on applicable compliance framework and internal customer policies this may impact technical choices.

AWS
EXPERT
answered 6 years ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.

Guidelines for Answering Questions