Watson Conversation Services and Virtual Assistant - Basic Summary
1. Basics - How Watson Conversation Service is able to provide meaningful responses to user questions?
Step 1
Collect a wide variety of “all possible”, end user questions or utterances
End User questions / utterances are tagged to: #Intent @Entity @Entity-Values @Alias
“When is the hotel pool open?”
#Intent=workinghours @Entity=HotelAmenity @Entity-Value=SwimmingPool @Alias=hotel pool
Above tag can be succinctly represented as “workinghours_HotelAmenity_SwimmingPool_hotel-pool”
Step 2
Map the above tag/s to suitable responses
Tag “workinghours_HotelAmenity_SwimmingPool_hotel-pool” or “workinghours_HotelAmenity_HotelPool”
Response: The Hotel Pool is open for use by customers from 11 am to 6 pm on all weekdays.
The actual mapping of the tags to the responses is done using dialog/chat flow
If the tags in Step 2 are very coarse, then the dialog / chat flow needs to be complex, to engage with end user and
provide final granular responses
If the tags in Step 2 are very fine grained, then the dialog / chat flow need not be complex, since our granular
mapping is good enough to figure out the exact response that needs to be provided.
3. Difference between Watson Conversation Service and a Virtual Assistant / Chat Bot
A Virtual Assistant or chat bot, usually means a ready-made and ready-to-use software which already has pre-baked
content as per suitable domain like banking, telecom, hotel industry etc. Typically a virtual assistant is hosted on the
cloud and is available as a SAAS. You can easily configure or customize many but not all aspects of the Virtual
Assistant like its name, the details of the answers that it responds back with and also the actual flow of dialog
between the VA and end user.
https://virtual-agent.watson.ibm.com/
Watson Conversation Service(WCS) is the underlying raw services and APIs, on which you can build a fully
customized virtual assistant for your specific business requirements. Typically, you will use WCS tools, api and
methodologies to:
Collect Questions - for the purpose of training the machine learning models of Watson Conversation service
Create Ground Truth - Ground Truth is built by taking collected questions and mapping them to the correct intents
Configure Dialog Component - The dialog that the user will use to communicate back and forth with the
Conversation Service is designed and built
Iterative Teach and Calibrate - The performance of the system is improved through modification of Ground Truth,
entities, and dialog flow