Spark supports scalable personalized search by training collaborative-filtering recommendations from user–item interactions. Its built-in ALS implementation learns latent features for users and products. section 9.3.2
A practical Spark recommendation pipeline:
- Convert clicks, add-to-cart events, and purchases into weighted implicit preferences. section 9.3.2
Weighted implicit preferences
A pipeline can assign clicks a weight of 1 while assigning add-to-cart and purchase events zero weight for a simple ALS setup. Those weights can be increased to represent stronger product interest. section 9.3.2
