Lecture 5 Flashcards
recommendation systems
help consumers find products that match well with their preferences
How are long tail and recommended systems related
consumers may be aware of some products but typically those are the ones available in the head of the tail
What type of data is in RS
Explicit data (ratings)
Implicit data
Explicit data
people are asked explicitly to rate items
types of recommendation systems
1.content-based
2.collaborative-filtering
content based-recommendation systems
making recommendations based on past actions
collaborative-filtering systems
based on activities and preferences of other users similar to you
Data in RS
Explicit and Implicit
how to build a profile of a user
E (attribute score * normalized rating )/ sum of attribute score
How to build recommendation system
by Cosine Similarity (the dot product for normalized vectors)
How does RS work?
work based on the similarity between either the content or the users who access the content
Interpreting results from Cosine Similarity
0 means no similarity, whereas 1 means a match
Problems with RS
- causality, because indeed prediction is being measured
- consider the counterfactual effect: what would have happen if the user not seen the recommendation
When the Recommendation system is accurate but useless
doing the groceries
RS will recommend bananas, but without the system I would buy them anyway.
The RS is 90% accurate but the incremental value is 0, will not change my behavior
Calculating content-based filtering
- Normalize rating
- Build a user profile
- Cosine similarity