Recommendation samples

The first demo of the Recommendation Engine widget is available as a preview here:

It is currently used for recommending user-generated content (videos) on a small test video sharing community dataset with 2000 videos and 500 users, a modest testbed for an ambitious project, but we believe our unique multi-faceted approach to recommendations will scale to much larger systems and datasets in the future.

To power the recommendations, the widget uses server-side calls to the
OpenRecommender Algorithm Web Service (currently running on Google's AppEngine with an implementation of Mahout,
GATE, JENA, and Drools).

More to come soon, including an even simpler, more generic HTML5/CSS3/jQuery widget which can easily be embedded on an external site or within any app that can support W3C Widgets. The goal of the widget is that it can be used to instantly recommend any type of item within the taxonomy (long-term) and ideally it should perform extremely well in aiding the following content choices for users (short-term):

• Books (based on authors or genres)
• Photos (based on photographer or types)
• Music (based on artists or songs played)
• TV (based on actors or shows viewed/DVR'd)
• Movies (based on director or IMDB voting history)
• News (based on topics or OPML sources of interest)
• Events (based on location or previously attended events)
• Products (based on category or price preferences)

*NOTE: recommendations can be generated for an individual person or a wider audience such as a group (or entire user-base of a given site), and will be capable of having personal-data input to improve their quality, or not!!! The choice will be in the hands of the implementer and their user community.

After the code behind the Recommendation Engine is at a point where we can test this (later this year), we'll launch it live and perform QA through invitation-only Beta tests, and, finally open source the first official version as OpenRecommender 1.0 through the community.