Reommendated system:
Key problems:
A.gathering known ratings for matrix
B.Extrapolate unknown ratings from the known ones.
C.Evaluate extrapolation methods
A.gathering known ratings for matrix
learn rating from user action
B.Extrapolate unknown ratings from the known ones
Key problem:
matrix is sparse./most people have not rated most items.
Three approaches to recommender systems:
1.content-based
2.Collaborative
3.Latent factor based
1.content-based
main idea:recommend items to customer x similar to previous items rated highly by x.
Example: recommend movies with same actor../recommend websites with similar content
Steps:
A.
for each item, create an item profile(is a set of features, such as movies :author, title, actor../text: set of important words in document)
In order to pick important features, using the TF-IDF
Doc profile: set of words with highest TF-IDF scores, together with their scores.
B.
Use the items to recommend .
Pros:
No need for data on other users / recommend to users with unique taste、new and unpopular items.
Cons:
Finding the appropriate feature is hard. Never recommends items outside user’s content profile.
2.Collaborative filtering
Steps:
Find set N of other users whose ratings are similar to x, then estimate x based on ratings of users in N.
Pros:
Work for any kind of item.
Cons:
Need enough users to find the match.
can not recommend items to someone with unique taste.
Tends to recommend popular items.
3.Latent factor based
C.Evaluate extrapolation methods
Using the known rating to compare the precision.