IMPORTANT:This module is temporarily abandoned in favor of Browsing History Recommender module. Please refer to . Future release of the module would take user relations from social networking modules and then make the user-to-user recommendation.
The basic idea of this module is to automatically find similar users in the community, and then recommend new contents to users based on the fact that other similar users liked the contents. This is actually an implementation of collaborative filtering used in many recommender systems.
To calculate similarity among users, we now only adopts the user-comment relationship. That is, if two users made comments to the same nodes for several times, then we consider them quite similar to each other. This is especially useful for communities that have forums enabled. There might be other approaches that calculate similarities between users, but I haven't thought about it thoroughly. You are welcome to submit your ideas to the issue queue.
The core algorithm that calculates similarities between users and makes predictions is provided by the Recommender API module. If you want to write your own cool modules involving collaborative filtering algorithms or vector-based algorithms such as PageRank or PCA, please check out the Recommender API module.