Download presentation
Presentation is loading. Please wait.
Published byAlexandrina Franklin Modified over 6 years ago
1
Chatbots Best Practices & Design Patterns…
@rsantrod @iamsoham
2
Design Patterns Best Practices Personas Flow-chart approach
Intent Separation Entity Definition Multi-language Error Handling Bot training Instant apps Channel agnostic design Personas Flow-chart approach Multi-bot approach Custom component approach Best Practices
3
Cloud & Mobile Solution Specialist
Oracle ACE Blogger ( & blog.avanttic.com) Cloud and Mobile Consultant UI Designer & Architect Oracle ACE Associate Tech Enthusiastic Blogger & Writter (
4
The architecture and functionality of the Oracle Mobile Cloud enterprise
Build RESTful APIs using Oracle ADF BC as REST, and with Node.js Deploy web services on Java Cloud Service and Application Container Cloud Service Integrate different systems and building Mobile ready APIs using Mobile Cloud An overview of Oracle JavaScript Extension Toolkit Build hybrid applications using Oracle JavaScript Extension Toolkit Create mobile-first Uis using zero- code platforms: Oracle VBCS and Oracle MAX Build a Chatbot for your enterprise using Oracle Intelligent Bots Cloud Service
5
Chatbot terms I want to request an appointment for tomorrow at 12
Utterance I want to request an appointment for tomorrow at 12 Intent Entity Introduce the terms we are going to talk about: Intent, utterances, entities. Maybe channel
6
Best Practices
7
Best Practices: Intent separation
Separate intents. Check out for utterances overlap Use batch testing when new intent is added Mixing of different languages within the same bot
8
Best Practices: Intent separation
Separate intents. Check out for utterances overlap Use batch testing when new intent is added Mixing of different languages within the same bot
9
Best Practices: Intent separation
Separate intents. Check out for utterances overlap Use batch testing when new intent is added Mixing of different languages within the same bot
10
Best Practices: Intent separation
11
Best Practices: Entity definition
System entities when possible Custom entities can be defined
12
Best Practices: Entity definition
Regex enties
13
Best Practices: Entity definition
I want a medium thin pepperoni pizza Size Crust Type Pizza Composite entities to define sub entities
14
Best Practices: Bot training
Normal training
15
Best Practices: Bot training
Batch training
16
Best Practices: Bot training
Batch training
17
Best Practices: Multi-language
Detect user’s language: translation services Option 1: translation services
18
Best Practices: Multi-language
Answer in user’s language Resource bundles
19
Best practices: Error handling
20
When you don’t need a chatbot
Mid/High Volume Search Activities Form-filling Login Command Driven Activities One doesn’t need chatbot. Think before you suggest chatbot as a solution to clients. It is innovative and cool but it is not supposed to solve all the problems in the world. Some of the things for which might not be the answer : Login activity Form filling activity Mid/High volume Search activity Command driven activities for example smart home systems,
21
Best Practices: Instant Apps
Use for things that chatbot is not meant for, i.e., login, filling forms etc. Use instant apps for these purposes.
22
Best Practices: Instant Apps
Use for things that chatbot is not meant for, i.e., login, filling forms etc. Use instant apps for these purposes.
23
Best Practices: Channel agnostic design
Dont use card layouts, global links, location components. (for example WeChat doesn’t support that) Specific properties in user object can vary from channel to channel, don’t design bot based on it.
24
Design Patterns
25
Bot – What & Who WHAT Pizza Home Automation Retail Leave/Timesheet
HR - Queries IT Service Desk WHO Age group Employee Public Channel Automation Bot
26
Bot – Personas – Know your User
Slightly older 3 kids Not so tech-savy Not a mulit-tasker ~Hardworker
27
Bot – Personas – Know your User
Slightly older 3 kids Not so tech-savy Not a mulit-tasker ~Hardworker Middle aged ~1 kid Average tech-savy Finds his way around Hardworking
28
Bot – Personas – Know your User
Slightly older 3 kids Not so tech-savy Not a mulit-tasker ~Hardworker Middle aged ~1 kid Average tech-savy Finds his way around Hardworking Young Extreme tech-savy Well informed Easily distracted Excited about new stuffs
29
Channel - Selection Txt Chat Channel Properties FB Messenger WeChat
> 1.2 Billion active users. WeChat > 1.1 Billion active users. In Web Most accessible In App custom mobile app Slack Enterprise messaing, Collaboration tool. Used extensively in DevOPs Skype 4 Business Enterprise messaging Alexa, Google Home Audio device Telegram Dominant chat in some markets. Text Most used and most accessible. Limited interaction UI. Texts costed. Txt
30
Design Patterns: Flow-chart approach
I have been scammed in marktplaats Someone broke into my house My car is stolen Intents FLOW When/How When/Where When/Where Entities Requirement refinement Bot’s boundary definition Intent identification Initial set of utterence definitions Entity identification Custom component requirements Flow overview definition File a report File a report File a report
31
Flow Chart Approach Advantages Requirement refinement
Bot’s boundary definition Intent identification Initial set of utterence definitions Entity identification API definitions Flow overview definition
32
Multi bot approach Respond in user’s preferred language
Keep dialog flow same Add languages easily
33
Multi-bot approach MyBot_nl Master Bot MyBot_es Detect Language
setVariable Reusability Develop-once Use everywhere Agile approach Extend your bot’s capabilities MyBot_de
34
Design Patterns: Custom component approach
Bot Flow is comparatively small Only NLP and Entity matching is used Strong integration with Backend Business logic intensive flow Flexibility in bot platform Wit.ai, Api.ai Intent Match Component A Component B Component C
35
Bot Maturity Model Level 1 Level 2 Level 3 Interaction Intelligence
API queries Simple Q&A One Language One Channel Word based rules Menu based Human to bot interaction Links for more information API transactions Multi channel Multi language Line based intelligence Training of NLP model Bot initiates conversation Event producing API intelligent queries Mood detection Context from channel State machine Human Handoff GEO Self learning Multi person Case Process interaction Bot-to-bot interaction Conversation intelligence B2B Conversation listening Historic analysis Interaction Intelligence Integration
36
Conclusions
37
Conclusions Separate Intents Define Entities
Use Batch/Run Report Regularly Use Translation Services and Resource Bundles Draw some flow diagrams Define the boundaries Take care of channel-agnostic features Keep the flow fexible
38
Questions www.linkedin.com/in/rsantrod
@rsantrod @iamsoham
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.