15
Conversational Apps – Episode 2: Training & Preparing Your Conversational System

Building with Watson - Training and Preparing Your Conversational System

Embed Size (px)

Citation preview

Page 1: Building with Watson - Training and Preparing Your Conversational System

Conversational Apps – Episode 2: Training & Preparing Your Conversational

System

Page 2: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

Housekeeping Items

• Webinar console widgets • Resources, Help, Q&A

• Live presentation will be followed by Q&A questions• Ask questions using chat window• Or on Twitter: @IBMWatson #BuildingWithWatson

©  2015  International  Business  Machines  Corporation

Page 3: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

Our Experts:

©  2015  International  Business  Machines  Corporation

Zach Walchuk - Developer Evangelist, IBM Watson Developer Cloud

Doug Ayers - Client Technical Specialist, IBM Watson

Page 4: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

Our Goals

• We want to give you all the tools you need to build a conversational application – an app that talks with your users

• Today, we'll cover the steps involved in training and preparing your conversational system

©  2015  International  Business  Machines  Corporation

Page 5: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

Natural Language Classifier

� The classifier “understands” what you're asking� E.g. “What is the address

of your school in Tucson?”, NLC intent => “Campus_Location”

Page 6: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

How to use Natural Language Classifier� Build a 2 column spreadsheet listing typical

questions paired with the corresponding class

� Create and train a classifier by calling Watson service� Simple web form on Watson Developer

Cloud� Reference the classifier in the dialog

configuration� Invoke from application in real time to classify

questions©  2015  International  Business  Machines  Corporation

Page 7: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

Dialog• Lets user's interact with app

using natural language- Simple question and

answer pattern or more contextual back and forth responses

• Uses NLC to understand intent and maps the question to the right answer

Page 8: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

How to use Dialog• Create the dialog “script” - an xml file• Configure conversation flow• Define entities and concepts• Examples and tutorials available on

Watson Developer Cloud• Use the Dialog Tool to iteratively build

and test• Invoke from application to start your

conversation©  2015  International  Business  Machines  Corporation

Page 9: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

Retrieve and Rank• Searches for best

answer for the infrequently asked questions not answered by dialog

Page 10: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

How to use Retrieve and Rank• Allocate a SOLR cluster (Watson service call)

• Edit a configuration file to define you data fields

• Create a collection and upload the documents from which answers will be retrieved (Watson service call)• Structured JSON file...not raw documents

• Create a file of “Ground Truth”• CSV file of sample search questions followed by relevancy

scores of documents

• Create the “Ranker” (run a script file from command line)• Rankers help insure the search service returns the most relevant

results first

• Test your search in a browser and invoke the search from your application

©  2015  International  Business  Machines  Corporation

Page 11: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

Are you ready to get started?Docs: - NLC, Dialog, R&R, Doc Conversion

Demos:- NLC, Dialog, R&R, Doc Conversion

See the services in use: - What's In Theaters and Watson Trend apps

Join our Slack channel for more discussion immediately following the webinar

©  2015  International  Business  Machines  Corporation

Page 12: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson

The Watson Developer Challengefor Conversational Apps

• An online hackathon with at-the-elbow support from our team

• First prize includes tickets to World of Watson and a chance to show off your app at the World of Watson hackathon

• We'll be kicking off soon...follow @IBMWatson for details!

©  2015  International  Business  Machines  Corporation

Page 13: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

Q&A

Page 14: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

What’s next?• Look out for a follow up email with a copy of these slides, a recording of the

webinar, Q&A recap, and additional resources

• The series continues bi-weekly on Wednesdays @ 12pm ET / 9am PT• Beyond  the  Basics  for  Conversational  Apps -­ Wednesday,  February  24• Deploying  Your  Watson  App  to  Bluemix -­ Wednesday,  March  9• Conversational  Application  Showcase -­ Wednesday,  March  23

Page 15: Building with Watson - Training and Preparing Your Conversational System

#BuildingWithWatson©  2015  International  Business  Machines  Corporation

Thank you for attending!

Contact usPhone: 1-877-253-0308

Email: [email protected]: Watson Developer Cloud