recommender systems
TRANSCRIPT
Recommender Systems
Anastasiia Kornilova
Agenda
Problems
Evaluation
Algorithms
General overview
We are overloaded of information:
• Books • Movies • News • Blogs • TV-channels • Music • …
As user:
• Do we need all of this things?
• No
• Can we choose most appropriate of them?
• Yes
• How?
• Recommender systems!
As Business Owner:
Do you need Recommender System?
• Netflix:
• 2/3 rented movies are from recommendation
• Google News
• 38 % more click-through are due to recommendation
• Amazon
• 35% sales are from recommendation
•Celma & Lamere, ISMIR 2007
Domains of recommendations
Content to Commerce
• Information
• News
• Restaurants
• Vendors
• TV-programs
• Courses in e-learning
• People
• Music playlists
One particularly interesting
property
• New items (movies, books, news, ..)
• Re-recommend old ones (groceries, music,…)
Examples: Retail
Examples: Banking
Deposits:
Deposit 1
Deposit 2
Credit products:
Credit card 1
Credit card 2
Personal loan 1
Personal loan 2
Insurance:
Endowment
Travel insurance
Service packages:
Premium package
Customer 2
Premium package
Travel insurance
Credit card 1
Customer 1
Travel insurance
Personal load 2
Deposit 1
Consultant can be replaced by Recommender System
Examples: Hotels
Examples: Advertisement
Shopping (browsing) history RSE
Purposes of Recommendation
Recommendations themselves (Sales, information)
Education of user/customer
Build a community of users/customers around products or content
Whose Opinion?
“Experts”
Ordinary “phoaks”
People like you
Personalization Level
Generic/Non-Personalized: everyone receives same recommendations
Demographic: matches a target group
Ephemeral: matches current activity
Persistent: matches long-term interests
Explicit input based RSE
Rating
Review Vote
Like
Implicit input
based RSE
Click
Purchase
Follow
Recommendation Algorithms
1. Non-Personalized Summary Statistics
2. Content-Based Filtering
Information Filtering
Knowledge-Based
3. Collaborative Filtering
User-user
Item-item
Dimensionality Reduction
4. Others
Critique / Interview Based Recommendations
Hybrid Techniques
Non-Personalized Recommender
Best-seller
Most popular
Trending Hot
Best-liked
People who X also Y
Personalized Recommender:
Collaborative Filtering
Use opinions of others to predict/recommend
User model – set of ratings
Item model – set of ratings
Common core: sparse matrix of ratings
Evaluation
Lift Cross-sales Up-sales
Conversions Accuracy Serendipity
Problems
“Cold start”
New user
New item
New system