Bits and pieces that might be useful for training a Watson Conversation service
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
training
LICENSE
README.md

README.md

conversation-starter

Bits and pieces that might be useful for training a Watson Conversation service. This is just an experiment to see if there are any reusable chunks that would speed up development of projects that use the Conversation service.

Training data

Intents

It seems likely that Conversation based chat bots will want some common elements. Since the easiest subset of a workspace to share are intents, due to the CSV import, those seemed like an obvious place to start. Plus there are already some overlaps in the sample projects out there to borrow from!

The following rough and ready jq command will reverse the CSV import process and extract intents from an exported workspace JSON file.

jq --raw-output '.intents[] | { "intent": (.intent), "example": .examples[].text} | [.example, .intent] | @csv' < exported-workspace.json > intents.csv

Template workspaces

For now there is just a basic starter workspace that just contains a common set of intents and an example conversation start dialog node to provide a simple starting point.

Possible additional workspaces:

  • small hello world example
  • chit chat example

I used the following jq command to remove some of the unnecessary content from exported workspaces ready for sharing:

jq 'recurse(.intents[]?, .examples[]?, .entities[]?, .dialog_nodes[]?, .values[]?) |= del(.created, .created_by, .metadata, .updated, .modified_by, .workspace_id)' < exported-workspace.json > workspace.json