PSALSA Example 3: User Similarity and Product Recommendation - Teradata Vantage

This example finds similar customers and recommended products for them.

Teradata recommends PSALSA for recommending products to similar users when the product line has limited pairwise history or changes frequently. An example of a product line that changes frequently is apparel, which is subject to seasons and trends.

Input

• vertices: user_product_nodes, which has customer names
• edges: women_apparel_log, which reflects customer shopping patterns

SQL Call

This call outputs a maximum of two similar users (hub) and recommends two products (authority) for each user, using frequency of purchase as a weight factor.

SELECT * FROM PSALSA (
  ON user_product_nodes AS vertices PARTITION BY nodename
  ON women_apparel_log AS edges PARTITION BY username
  USING
  SourceKey ('username')
  TargetKey ('product')
  EdgeWeight ('frequency')
  MaxHubNum (2)
  MaxAuthorityNum (2)
  TeleportProb (0.15)
  RandomWalkLength (500)
) AS dt ORDER BY username, hub_score DESC, authority_score DESC;

Output

The output shows possible recommendations, based on hub_score and authority_score. For example, the seller can recommend pajamas to Sandra and Susan because they and Sally have similar scores. The hub_score and authority_score values vary with every run.