Recommendation Task

Preference Recommendation

• Given a collection of (user, item) interactions, predict an unknown rating by a user u to an item i

Sequence Recommendation

• Given a history of interactions, predict what user u will choose next

Utility Matrix

• Expresses relation between users and items
• An element of the matrix represents a user’s preference toward an item
• Goal of recommender system is to fill in the blank entries
• it is not necessary to predict every blank entry in a utility matrix
• is is often desirable to discover entries that are likely to be high rated

Key Step

1. Gathering “known” ratings from matrix
2. Extrapolate unknown ratings from the known ones
3. Evaluating extrapolation methods

Types of Recommender System

• Content-based
  • Focus on properties of items
  • Similarity of items is determined by the similarity in their properties