+

Search Tips   |   Advanced Search

LikeMinds Recommendations

Personalization contains a dynamic recommendation system based on LikeMinds. LikeMinds is software used with the e-commerce applications. LikeMinds analyzes user interactions that occur on the Web site and generates real time predictions and recommendations to the Web site users.

Real time predictions are generated by three LikeMinds engines using recommendation rules within Personalization. These rules, called recommend content, base their predictions on transactions logged through Personalization's rating and action beans.

When a user visits the Web site, rating and action beans log captured transactional data. If the e-commerce Web site is set up so that users can rate content (or products), we use Rating beans to capture rating data. Similarly, if we use shopping cart technology, we use action logging beans to capture content affinity behavior to capture shopping activity. Both rating and action data is stored in the database. For example, the following types of transactions may be recorded:

Use recommend content rules, LikeMinds surfaces results through has a set of recommendation engines. These engines predict relevant content for users based on their past Web browsing habits.

Typically, after a user has rated a minimum number of items or completed a minimum number of transaction activities, that user is assigned has a set of mentors. A mentor is a specially designated user who has visited the e-commerce application a number of times, and whose profile is similar to the user's. LikeMinds uses a technique called collaborative filtering to build a mentor's profile for each user to predict how much a user will like particular items and which items that user will enjoy, buy, or add to their shopping cart.

Predicting a matching product to go with a user's selected product, independent of actual user preferences, is accomplished by the discovery of probable pairs of product matches to be recommended. This concept is called item affinity and uses a family of algorithms different from collaborative filtering. While collaborative filtering uses its algorithms to discern the highly variable affinities between individual Web-surfers, the item affinity approach looks at relationships that can exist between items.

We can use LikeMinds in a variety of situations, including:


Parent: Personalization

Previous topic: The Portal User resource collection

Next topic: Feedback and analytics