Support Questions

Find answers, ask questions, and share your expertise
Announcements
Celebrating as our community reaches 100,000 members! Thank you!

Oryx ALS Collaborative filtering essentials

avatar
Explorer

Hi,

I need help with understanding basics of Oryx ALS Collaborative filtering essentials.

 

What exactly means "features", for example each user has 10 features and also each item has 10 features, some of them positive, some negative numbers. First I thought it represents matrix user-item, but I have more than 10 users and more than 10items.

 

When ingesting or learning new model, there is by default weight set to 1. Imagine I have various rating of movies, where 1 represents bad rating and 5 great rating. What weight should I set? I have tried to undestrand source code from which it seemed to me that weight should be in interval (0,1).

 

Thank you for help!

1 ACCEPTED SOLUTION

avatar
Master Collaborator

Features means the number of latent features in the factored matrix model. If the user-item matrix A is factored as A ~= X Y', then the number of features f is the number of columsn of X and Y.

 

Weights are not ratings. No, weights can be any value. One approach is to view any interaction at all as a "1". You might instead treat bad ratings and negative weights, and good ratings as positive weights.

View solution in original post

2 REPLIES 2

avatar
Master Collaborator

Features means the number of latent features in the factored matrix model. If the user-item matrix A is factored as A ~= X Y', then the number of features f is the number of columsn of X and Y.

 

Weights are not ratings. No, weights can be any value. One approach is to view any interaction at all as a "1". You might instead treat bad ratings and negative weights, and good ratings as positive weights.

avatar
Explorer

Hi,

 

thank you for quick reply.

 

I am quite new in recommendation domain. What exactly mean latent features?