Chipmark Development Blog

For our latest release, we implemented a chipmark recommendation system. In no way are the recommendations used to promote other websites or products. The recommendations are generated using an algorithm that examines a user’s existing chipmarks against every other user’s chipmarks. The algorithm was borrowed from the GroupLens research, also from the University of Minnesota. The recommendation system is run once every week, and it takes approximately six hours to complete.

When the thread is awoken to generate recommendations, a script is run that exports each user and their chipmarks from the database to a flat file. A parser then reads that file into memory. Once the data is in memory, the data is divided up between four threads. The chipmarks of user a are compared to the chipmarks of users b, c, d, etc. The recommender finds the user that has the highest number of chipmarks in common with user a, we’ll call him user x. The system then finds five chipmarks that user x has that user a does not have. Those five chipmarks will be recommended for user a. This repeats for each user.

Any feedback on how relevant these recommendations are would be appreciated. Feel free to examine the source code. The relevant code is located in the bookmarks.recommender package.

This entry was posted on Saturday, May 5th, 2007 at 9:40 am and is filed under Announcements. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.