SEPTEMBER 2016 UPDATE: 

I wrote some R Hooks to talk with Conversation service - here:

https://github.com/rustyoldrake/R_Scripts_for_Watson/blob/master/Watson-CONVERSATION-Code-Backbone.R

_______________________________________

IBM Watson Launched CONVERSATION Service July 2016 - - here is what I found on my 1st tour....

Step 1 - Go To Bluemix https://console.ng.bluemix.net/ and select USE SERVICES OR APIs  (signing up to free account takes 2m)

here is bluemix dashboard.  click bottom right box to launch a service

Step 2 - Select CONVERSATION SERVICE tile

new tile!  Conversation.  It's a mashup of NLC, Dialog and other services.  with terrific UX/tooling

Step 3 - Think of a nice name and CREATE  - you can leave it unbound.

creating the service in bluemix.  Takes about 10-20 seconds to be available.

Step 4 - Select LAUNCH TOOL (bottom right)

this is the step I normally go to "Service Credentials" and start hammering service.  For now, as of July 2016 - the best way to create the Conversation Service is with Tooling.  Launch Tool! 

Step 5 - New tooling will launch - select CREATE.  Pick a name and some simple example of Chat Bot

my example was a basic intent detector for coffee, cheese, chocolate and wine.  Yum!

Step 6 - Start with basic INTENTS - Add a 3 or 4 intents to test - and give service 5-10 examples of user inputs that should MAP to intents.  (Classes) - this may look familiar to NL Classifier users. 

see NLC and Conversation documentation for details.  Links below.

Step 7 - Click on CREATE DIALOG.  Then start playing with nodes.  FIRST NODE - CREATE "conversation_start" and create new CONDITION (Starting node) - see docs for this.

experiment here - you can test real time, so you'll get hang of basic flow pretty quick

Step 8 - First box is prompt.  Then build a decision tree from your INTENTS like below #coffee and #chocolate . Also LOOP back (continue from -> Go to condition)

Also - create a LOOP back to start for testing "Continue From should click to "go to response" in the first conversation node (i.e. reset the talk once it provides answer)

Step 9 -  Try it out!  Test using icon top right corner

Step 10 - FUTURE - Add downstream functionality - I have some NL CLassifier Ground truths (Coffee, Cheese, Wine) that I'll 'wire in' - so Bot can ask follow up question and then recommend a specific delicacy.

Summary -  so in about 30 minutes of playing around we....

1. Created a Bluemix Service called Conversation

2. Launched Tooling

3. Built some simple INTENT detectors

4. Did a logic flow in Dialog

5. Tested a simple bot to detect whether I was interested in Coffee, Wine, Chocolate or Cheese.

I had a bit of a learning curve in Dialog area, so may blog more on that later.  Also, I am keen to plug my Code into this (hit APIs) so watch this space.

SUPER SIMPLE & CLEAR 2 Minute Video

Other Conversation Links:

• http://www.ibm.com/watson/developercloud/conversation.html
• https://watson-api-explorer.mybluemix.net/apis/conversation-v1
• https://github.com/rustyoldrake/R_Scripts_for_Watson/blob/master/Watson-CONVERSATION-Code-Snippet.R (Very early code)
• coffee NL Classifier - https://dreamtolearn.com/ryan/r_journey_to_watson/23
• wine NL Classifier - https://dreamtolearn.com/ryan/r_journey_to_watson/24
• Zach Walchuk Customer Service Chat-Bot https://www.linkedin.com/pulse/super-simple-customer-service-chat-bot-you-can-build-ten-zach-walchuk

Build a chatbot in 6 minutes with IBM Watson Conversation

=======

Appendix A - here is an example of curl/command line syntax

SENT - CURL:

Ryans-MacBook-Pro-2:~ ryan$

curl -X POST -u e63f524d-b6f4-9999-905b-e6bXXXe99b87:KQ4d0EggCyGR -H "Content-Type:application/json" -d "{\"input\": {\"text\": \"hot coffee\"}, \"context\": {\"system\": {\"dialog_stack\": [\"root\"], \"dialog_turn_counter\": 1, \"dialog_request_counter\": 1}}}" "https://gateway.watsonplatform.net/conversation/api/v1/workspaces/2ded4293-871d-4b24-XXXX-4c8b1289be81/message?version=2016-07-11"
 

RESPONSE:

{error: no user with username '"intents":[{"intent":"coffee","confidence":0.9871662023563588' was found}],"entities":[],"input":{"text":"hot coffee"},"output":{"log_messages":[],"text":["I'm sorry. I don't understand. Please try again"],"nodes_visited":["node_9_1468894011999"]},"context":{error: no user with username '"system":{error: no user with username '"dialog_stack":["root"],"dialog_turn_counter":2,"dialog_request_counter":2},"conversation_id":"1f84342f-5aad-XXXX-b40f-dd1XXXb66b4f"' was found}' was found}

=======

Appendix B - here is an example of R syntax:

library(RCurl) # install.packages("RCurl") # if the package is not already installed
library(httr)
library(XML)

url_CON="https://gateway.watsonplatform.net/conversation/api/v1"
version="?version=2016-07-11"
username_CON <- "e63f524d-xxxx-xxxx-905b-e6b1d4e99b87"
password_CON <- "KQ4dxxxxxyGR"
workspace_CON <- "xxxxxxxx-871d-xxxx-xxxx-4c8b1289be81"

## testing
response <- POST(url=paste(url_CON,"/workspaces/",workspace_CON,"/message",version,sep=""),
                 authenticate(username_CON,password_CON),
                 add_headers("Content-Type"="application/json"),
                 body = '{ "input": { "text": "coffee cake" } }',
                 encode = "json"
                  )

response
response_text <- content(response, "text", encoding = "UTF-8")  # or encoding = "ISO-8859-1"
response_text

##########

> response_text <- content(response, "text", encoding = "UTF-8")  # or encoding = "ISO-8859-1"
> response_text
[1] "{error: no user with username '\"intents\":[{\"intent\":\"coffee\",\"confidence\":0.9871662023563588' was found}],\"entities\":[],\"input\":{\"text\":\"coffee cake\"},\"output\":{\"log_messages\":[],\"text\":[\"Welcome to the Tasty Helper!  \\nWhat are you looking for?\"],\"nodes_visited\":[\"node_6_146889888885\"]},\"context\":{error: no user with username '\"conversation_id\":\"28c696cd-5694-4377-99a5-688888d3ab266\",\"system\":{error: no user with username '\"dialog_stack\":[\"node_6_14688888593485\"],\"dialog_turn_counter\":1,\"dialog_request_counter\":1}' was found}' was found}"

=====

https://developer.ibm.com/answers/topics/watson-conversation.html

https://ibm.biz/BdrVb7

1 Comments

• Public Comments

Before you can comment, you need to sign-up or login

Jake Madden

responded August 2016

I played around with this service the other night. So far I think the wit.ai training for conversations is much more user friendly. The Watson speech to text is much better, though. The conversation setup for the watson side was not working as expected and I couldn't find much in the way of debugging what was happening. I will keep experimenting.

---