Eric Jernigan, Manager of Application Development ... Interactive.pdf · NATURAL LANGUAGE PROCESSOR...
Transcript of Eric Jernigan, Manager of Application Development ... Interactive.pdf · NATURAL LANGUAGE PROCESSOR...
BOTS!
Eric Jernigan, Manager of Application Development
Mississippi Interactive
RISE OF THE BOTS
Over 4 billion users around the world are using
instant messaging platforms
6 out of the top 10 apps used globally are
messaging apps
Messaging has now overtaken social networking
applications
POTENTIAL USE CASES
• Troubleshooting
• New user on-boarding
• Response suggestion
Customer Service Purchasing Help Desk Employee Support
• Cross Selling
• Product Suggestions
• Improving search
• FAQ
• Account Lookup
• Tracking
• Statistics
• App Insights
• Vacation time
• Meeting room booking
• Routine tasks
WHAT MAKES A GREAT BOT?
Easily solves user’s problem (and efficiently)
Better than alternative experiences
Runs where ever the user is
Natural / Intuitive to use
BOTS ARE PEOPLE TOO
Create a personality around your bot.
Mimic ways that people actually talk to each other.
Build experiences when interfacing with customers.
What is your Bots job?
HOW TO MAKE A BOT
Bot Frameworks
• RASA Core
• Dexter
• Microsoft Bot Framework
Build one bot
Host one bot
Maintain one bot
Target multiple channels
BUILDING A CONVERSATION
Build a flow diagram to test your conversation
A core principle in Bot design are Intents and Entities
These are parts of a conversation which encapsulates behavior similarly to screens in a traditional UI
REPRESENTATIONS VS UNDERSTANDING
Sentiment analysis / Machine translation
“king is to man as queen is to _____”
When AI can’t determine what “it” refers to in a sentence, it’s hard to believe that it will take over the world.
NATURAL LANGUAGE PROCESSOR (NLP)
NLP (Natural language processing) is the science of extracting the intention of text and relevant information from text.
Taking raw human language and return the basic parts of speech, nouns, verbs, dates, sentence structures.
NLP Platforms help developers create language capabilities in the least amount of time possible.
NATURAL LANGUAGE PROCESSOR (NLP)
Some popular NLP as a service platforms are:
1. LUIS.ai — By Microsoft
2. ASK (Alexa Skills Kit) - Amazon
3. Api.ai / Dialogflow— By Google
4. Watson — By IBM
5. RASA NLU
INTENTS
Intents are the intentions of the end-user and can typically be placed into 2 categories
1. Casual Intents
2. Business Intents
CASUAL INTENTS
Casual Intents can be defined as “Small Talk”
Casual Intents can also comprise of both positive and negative affirmations
Greeting Positive Negative
• Hi
• Bye
• Ok
• Yes Please
• No
• You stupid human
BUSINESS INTENTS
Intents that directly map to the business of the bot
• What is the score of the Broncos game?
When naming your intent try and use a understandable name
• GetScoreByTeamIntent
ENTITIES
The “metadata” of your Business intent
What’s the score of the Broncos game?
“Broncos” are the team for which the user “intends” to find the “score”
The process of finding entities can be understood as Part of Sentence (POS) tagging.
Like Intent naming. Take care of labeling and identifying your entities.
COMPOSITE ENTITIES
Having more than one entity inside it
For ex: Find me a Red Ferrari Convertible for sale
Composite: CarDetail
Intent - SearchCarIntent
Entities:
Make: Ferrari
Color: RedComponent:
Model: Convertible
TRAINING
3 Simple Rules for Intent and Entity Training:
1. You can never train it enough. Get more data, train more data
2. You can never train it enough. Get more data, train more data
3. You can never train it enough. Get more data, train more data
TRAINING
Ideally one should train the NLP with some real data.
• Great place to use existing chat logs, Facebook posts, Slack conversations.
Manufactured utterances are ok to bootstrap.
what was the release year of the movie Frozen
in which year was Top Gun was released
when did Star Wars came ( Poor English is sometimes encouraged )
who was that guy in the movie Braveheart
LEARNING PHASES
Supervised
Unsupervised
Supervised
Unsupervised
THE FUTURE OF BOTS
• Your speech will be your operating system
• The line between online and offline will blur
• We will rely less on traditional search engines