OpenRecommender's Re-launch in DRUPAL (July 1st, 2008-July 1st, 2010)

Welcome to the new home of OpenRecommender, an initiative to create the world's leading open source Recommendation Engine.

The project initially started in July 2008 as a single PHP file for generating recommendations for online videos, which would then be displayed within the JW Media Player in the recommendation post-roll that was introduced in version 3.16 of the player. The post-roll was powered mostly by a single ActionScript file which was optionally activated on the video's end event as a video reached the final second of its duration (i.e. ticked down to 0:00 remaining in the countdown timer).

After this initial investigation it was decided that a Recommendation plugin that supported additional triggers, behaviors and presentational styles would be beneficial. Luckily, by the time any interest was renewed in a larger full-blown Recommendation Engine, the JW Media Player had released its version 5.5 player which enables a much more robust plugin-mechanism, whereby plugins can now be Actionscript or Javascript-powered. The door is now open for open web technologies to take a place in the world's most excellent open source/commercial and widely used media player, and, with multimedia at the heart of the user-experience on the World Wide Web, the time has never been better for OpenRecommender to shine.

Most significantly, though, this initial investigation into recommending mostly user-generated video content (and a handful of mainstream media videos such as movie trailers, tv commercials, etc) was the beginning of a vision for a much larger mechanism for recommending content and items of all types, to people from all over the world.

Ideally, this community aims to build a Recommendation Engine capable of recommending just that any item, to any person, in a highly relevant manner, while requiring minimal inputs and thus maximizing the user-experience without sacrificing users' privacy. Any optional user-driven recommendations data would only be generated and/or shared with express consent of the users themselves.

While this may seem like a bit naiive nonesense to some, as researchers may be quick to cite the numerous attempts at intelligent and adaptive information retrieval (IR) systems which have been undertaken before with varying degrees of success (but usually coming up short when broad in focus), that is precisely where this project promises to pick up the slack.

None of the various initiatives undertaken (to this author's knowledge), pitted recommendation algorithms against one another in a sort of battle-royale for dominance of a particular category of recommenation. Most initiatives were funded by big research organizations who kept tight control over any code and/or intellectual property divulged via whitepapers, since Recomender Systems are a promising and highly patentable area of Computer Science research. With information so guarded, it is difficult for smaller projects to succeed, and for knowledge in the area of Recommender Systems to be disseminated widely.

Thus, OpenRecommender begins a break from this lab-centric approach by inviting the entire open source community to contribute their knowlegde by analyzing the existing system's code to find/fix faults, or, to augment OpenRecommender with as many new (GPL-compatible) algorithms as possible.

OpenRecommender's recommender system itself will be based on a combination of the best techniques for automated Recommendation generation, from basic concepts like:

• Roaming Profiles and
• Virtual Communities

to more complex ones, such as:

• Transparent
• Analytic,
• Taxonomic, and
• Categoric

Trust Hierarchies.
These are the main issues around fairly and equitably monitoring and monetizing communities online.
However, there are many other unsolved issues; for example how should the recommendation of one member of the community affect the rest of the population and their future recommendations. Many techniques exist for approaching these types of community problems, from member indexing techniques like:

• Collaborative Filtering,

to more comprehensive

• Behavioral Targeting

Finally, an even greater problem exists… what about the many MANY (vast majority, we dare say) sites who do not enjoy the millions of daily visitors that popular silo sites like Apple’s iTunes Store, Amazon, Yahoo!, YouTube, MSN and even Google enjoy to themselves. There needs to be a way to take empty data structures on the web and populate them with the knowledge already accrued at other sites over time. This is leading to an entirely new movement on the web based on a vision of a better, more intelligent web that uses personal agents to gather relevant information. There are many complexities to realizing this kind of full-fledged Recommendation Engine, and some of the foremost techniques for dealing with such complexities include everything from:

• Contextual Matching,
• Perceptual Mapping, and
• Mind Mapping

to

• Semantic Web agents,
• Semantic memory building, and
• Machine Learning technologies based on statistical inference.

Surely the solution to these many problems involve many pieces of information, from interdisciplinary studies spanning the gamut of the human intellect. This is why we invite ANYONE and EVERYONE to join the OpenRecommender project. Whether you are in the field of Psychology or Computer Science, Philosophy or Engineering, Arts or Information Technology, Business or Bio-Informatics, the creator of this project strongly believes that everyone can contribute to creating the world’s leading open source Recommendation Engine that will help all of us continually improve and simplify our digital lives.
Stay tuned over the next few months, as we will soon be launching our first set of Ontologies. The first ontology will be the MobileTV ontology, which is intended to describe content services distributed to wireless and mobile handheld devices. This is sure to be a controversial first release since so much corporate attention and venture capital has been invested in this area.
We hope many will come back regularly to contribute to this and any other future ontologies released through the OpenRecommender project. You can track and/or contribute to this project by signing up (HERE, or by clicking Register in the menu on the right).