Video and slides synchronized, mp3 and slide download available at URL http://bit.ly/2xbyEiA.
Robert Enyedi talks the ebay ShopBot, a personal shopping assistant available as a Facebook Messenger bot. He talks about the internals of the ebay ShopBot and how the processing of text, image and speech flows through the purpose-built system leveraging the product knowledge graph, commerce-aware input recognizers and the Dialog Manager. Filmed at qconnewyork.com.
Robert Enyedi works as a MTS Software Engineer at eBay.
2. InfoQ.com: News & Community Site
• 750,000 unique visitors/month
• Published in 4 languages (English, Chinese, Japanese and Brazilian
Portuguese)
• Post content from our QCon conferences
• News 15-20 / week
• Articles 3-4 / week
• Presentations (videos) 12-15 / week
• Interviews 2-3 / week
• Books 1 / month
Watch the video with slide
synchronization on InfoQ.com!
https://www.infoq.com/presentations/
ebay-shopbot
3. Presented at QCon New York
www.qconnewyork.com
Purpose of QCon
- to empower software development by facilitating the spread of
knowledge and innovation
Strategy
- practitioner-driven conference designed for YOU: influencers of
change and innovation in your teams
- speakers and topics driving the evolution and innovation
- connecting and catalyzing the influencers and innovators
Highlights
- attended by more than 12,000 delegates since 2007
- held in 9 cities worldwide
4. eBay Marketplace
at a Glance
2
One of the world’s largest and
most vibrant marketplaces
Data as of Q1 2017
12M
New listings
added via
mobile per week
$20B
GMV
1.1B
Live listings
67%
Transactions
that ship for
free (US, UK, DE)
60%
Platform GMV
touched by
mobile
80%
Items sold as new
5. 2017 Fun Facts: Velocity Stats by Region Frequency of product purchases via
desktop and mobile
US
11 sec
5 sec
4 sec
UK
1 sec
3 sec
8 sec
DE AU
16 sec
3 sec
18 sec
14 sec
26 sec
4 sec
6. The Major Trends in Commerce
4
AI is having a profound
impact on commerce
Users expect simple
and personalized
interactions
17. System Design Constraints
• User input:
– Minimal
– Fragmented through multiple interactions
– Limited forms
• Needs large amounts of data to accurately understand the input
• System output:
– Has to be user-friendly
– Limited mechanisms
15
18. Features of Third-Party Bot Frameworks
•Natural Language Understanding and Processing
–Tuned for general purpose bots
–Intent detection: (e.g. weather, flight scheduling, shopping)
–Entity extraction: (e.g. number, temperature)
•Limited non-linear conversation support
•Coarse grained bot memory
•In various states of availability and maturity of API, tools and
implementation
16
19. Why Implement Our Own Bot Back-end?
•Commerce-aware input analysis tuned with
eBay-scale data
•Fine-grained bot memory with long-term
secure storage
•Support for non-linear multi-turn dialog
•Full data from input analysis
17
20. ShopBot AI System
System Architecture
18
Bot Memory
ShopBot
Fabric
Dialog Manager
Input Processors
Facebook
Messenger
Knowledge
Graph
21. ShopBot Fabric layer with Facebook Messenger Integration
19
ShopBot Fabric
Facebook
Messenger
Webhook
Callback
Web View
Webhook
Send API
24. Natural Language Understanding
22
My husband needs some new black leather dress shoes but I want to
spend less than $80, what do you have?
Relationship: Husband
Gender: Male
Dominant Object: Dress shoes
Product Category: Men's Dress Shoes
Color: Black
Material: Leather
Max price: $80
Natural
Language
Understanding
33. Architectural Benefits
•Extensibility and Flexibility
–Integrations
–Search providers
–Knowledge sources
•Improved scalability
–Auto-scalable chat bot system
–Horizontal and vertical
•Depth of understanding
–Leverages the richness of eBay data
31
ShopBot AI System
Bot Memory
ShopBot
Fabric
Dialog Manager
Input Processors
Facebook
Messenger
Knowledge
Graph
34. Learnings
•Many users see ShopBot as a search engine: “shoes size 12”
•Handling unrestricted user input is challenging when specialized on
shopping:
–Users greet the bot
–Some expect it to know the weather
–Others use free text such as ”I don’t like those”
•Accurate response tops a fast response
•Ask the user for disambiguation
32
35. Status
•Work started in Jan 2016
•Released as public beta in Oct 2016
•A platform for collecting data and learning
•Full eBay inventory in the cloud
•Continuing innovation into non-linear dialog interactions
•Researching how to leverage AI in conversational commerce
to make shopping easier/faster/delightful
33