How LikeMinds generates recommendations

When a user logs on and navigates through your Web site, LikeMinds follows these steps to generate recommendations for that user:

  1. WebSphere Commerce creates a record for each new user in the Lps_User_Data table.
  2. WebSphere Commerce logs data for that user as that user navigates your Web site. The Lps_User_Trx table stores the user's clickstream and purchase behavior.
  3. WebSphere Commerce can then query LikeMinds for recommendations. Recommendations are surfaced by the recommend content rule. To receive Clickstream recommendations, WebSphere Commerce must record users' clickstream behaviors (that is, product detail views, shopping basket inserts, and so on).
  4. For a new user, if WebSphere Commerce queries LikeMinds for recommendations before mentors have been assigned, the server will assign mentors from a cached pool of mentors. If the server is unable to, for lack of profile data, match cached mentors to this user, the server will provide an empty set of recommendations. An important distinction; profile data means the transaction data for the current user and not the attributes of that user.
  5. Once a user's profile is stored in the database, the sifter utility can calculate mentors for that user. The sifter is a background utility which assigns a set of mentors to each user.
    • Mentor assignments are specific to each type of data.
    • Mentor assignments are stored in the mentor table associated with this type of data.
  6. As new transaction data is recorded for a user, the user is prioritized for reprocessing by the sifter to calculate new mentor assignments. Users are prioritized by a calculated 'sift priority', reflecting the percentage of new or changed profile data for that visitor.
  7. When WebSphere Commerce runs LikeMinds rules, LikeMinds looks up that user's mentors, and calculates recommendations.

Related concepts

(C) Copyright IBM Corporation 1996, 2006. All Rights Reserved.