Collaborative Filtering Manual A person who sends emails Automatic Historically U. Minesota GroupLens Project MIT FireFly Music BellCore Video recommender
Movie Lens Ratings are stored in database. Correlations for users are calculated.
Interfaces of recommenders: Filtering Recommender Prediction
Java Maven LensKit InteliJ
Analytical Framework for analysis of recommendation systems
- What are we recommending?
Text? Products or bundles? Other people? Sequences?
New items or re-recommending old ones.
Purpose of recommendation?
- Communitity building
ReferalWeb recomends another user in one's social graph based on request.
What the user is doing at the time of recommendation?
- Ordinary people
People like me
- Based on demographics
- Matching on current activity
Matching long-term interests
Privacy and Trustworthiness
Who wants to know about me?
Is the recommendation honest?
Types of output
- Types of input