LikeMinds Recommendation Engine architecture

LikeMinds bases its construction of predictions on user actions such as...

• History of the user's navigation
• Products that go with a product already selected by the user
• Products purchased
• Items added to or removed from a shopping basket

User preferences are collected by the engine and used to identify people most like the current user. People with a behavior similar to the current user become mentors for that user, with the LikeMinds Recommendation Engine assigning a numeric weight to each mentor based on the level of similarity.

The Recommendation Engine assembles a set of recommendations by finding the products each mentor recommends and creating a prediction vector containing the predicted rating of each product. With each predicted rating, it also stores a numeric value representing the confidence for the rating.

The confidence values determine the quality of predictions. The Recommendation Engine assigns a confidence level to each recommendation based on how many users have rated the recommended item and how similar the ratings are to each other.

A user is assigned a set of mentors only after he has rated a minimum number of items or completed a minimum number of transaction activities. We can configure probable pairs of product matches to be recommended (item affinity).

Specific engines must be configured, including the Preference Engine or Item Affinity Engine.

Parent topic: LikeMinds Recommendations