1

I am finding this task very challenging and unable to find an approach.

Given:Structred Dataframe of FinanacialReport Finanacial Report

I want to identify the intents given a text query. To be more precise: If I'm given a text query "What is Fiscal 2013 of Latin America?"

Output:388.00

The challenge I'm facing is it can be done for unstructured text/paragraphs suing NER? But how to proceed when given a dataframe like this?

Any help would be appreciated.

Nazim Kerimbekov
  • 4,712
  • 8
  • 34
  • 58
Soyam Maharathy
  • 73
  • 8

1 Answers1

1

I would approach it as follows (using Rasa):

1) Provide NLU training data where you tag your required entities. E.g. an entity location and a entity metric. This probably best done with the crf component. Maybe it's worth combining it with lookup tables.

2) Run a custom action action_query_financial_data when the intent for the text query is classified

3) In the custom action, use the extracted entities to query the dataframe.

Tobias
  • 1,880
  • 11
  • 17