How LikeMinds generates recommendations

When a user and navigates through the Web site, LikeMinds follows these steps to generate recommendations for the user:

1. Rating beans and action logging beans create a record for new users in the Lps_User_Data table, storing the user's resource ID, a user ID, and the number of items the user has rated or selected.

2. The beans log data for the user as the user navigates the Web site.

   The profile data is first stored in the server's cache, then the server writes all the new data to the database. The Lps_User_Rating table stores the user's explicit preferences. The Lps_User_Trx table stores the user's clickstream, purchase, and item affinity data.

3. The application queries LikeMinds for recommendations.

   • Preference recommendations are surfaced by the recommend content rule.

   • Clickstream recommendations (product detail views, shopping basket inserts) are surfaced by the recommend content rule.

   • Item Affinity recommendations are in the form:
     "This mouth guard is a likely product to go along with the hockey puck the user has just added to his or her shopping cart"

     To receive Item Affinity recommendations, the application must record users' likely product pair matches. In other words, the application must capture the current "content/product" context using the Action bean, in order to return those items most associated with that "content/product".

4. Depending on the engine we are using, the following step occurs next:

   Preference and Clickstream engines

   For a new user, if the application queries LikeMinds for recommendations before mentors have been assigned for that type of data (that is, Preference, Clickstream, or Item Affinity), 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.

   Item Affinity engine

   If the application is predicting item affinity product pair matches, it will collect data based on a set of definitions created, called an item affinity set. For input data, the item affinity set uses transactions from a specified input table.

5. Depending on the engine, the following step occurs next:

   Preference and Clickstream engines	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.
   Item Affinity engine	The accumulator generates item affinity product pair matches by analyzing data in the item affinity set (same as transaction table) and recording its findings to an output table.

6. Depending on the engine, the next step is as follows:

   Preference and Clickstream engines	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.
   Item Affinity engine	As new product selection behaviors are recorded in the transaction input table specified in the item affinity set definition, the accumulator uses this data to calculate new item affinity recommendations.

7. When the application runs LikeMinds rules, the following occurs, depending on the engine:
   • Preference and Clickstream engines: LikeMinds looks up that user's mentors, and calculates recommendations.
   • Item Affinity engine: LikeMinds calculates the most likely item-content pairs based on accumulated item transaction history.

Parent topic: LikeMinds Recommendations