Presenters: Stephen Green, Sun; Paul Lamere, Sun
The first half of this presentation focused on explaining the Long Tail of marketing and specifically the music industry. A lot of examples were shared about recommendation systems and solutions that fall short in some areas, sometimes. All in all, Stephen and Paul were big fans of the Amazon recommendation system, but because it is the most visible system it is easy to pick on and point out faults when they occur. The most humorous of which was a Pat Robertson book which suggested that people interested in it would also be interested in a book on unlocking the secrets of the gay sex life. How could something like this happen? Amazon formerly weighted suggestions based on recent trends of user interests (think of the end of the tail). A coordinated group of users would be able to significantly influence the recommendation engine and thus, cause something like this. Amazon has since changed this algorithm, but the illustration was useful.
The latter half of the session went into more detail on how Project Aura is able to achieve really good recommendation results for the average implementer with very little effort. Leveraging OpenID and Attention Profile Markup Language(APML), the data is open and accessible encouraging consumer ownership. There was a BIG gotcha at the end of the presentation that caused the level of enthusiasm in the room to burst and leak out the doors rapidly. Project Aura is a hosted web service model. The preference data gathered is all shared with the community. Donation required for use of the services is the data. This will make this solution a no-go for a lot of organizations but would still be a viable option for consumer based startups that are looking for a quick recommendation engine. It is based on Project Caroline for scalability. There are some significant limitations about customization but for a quick startup it is still a great project.
The project is not available as of yet, September is the current target but taste.sourceforge.net is a good one to check out for something today.
You can read more information about the project on Paul's Blog - http://blogs.sun.com/plamere/entry/recommendation_for_the_rest_of
-mw
1 comment:
Thanks for the post. One note - Project Aura will be released entirely as open source, so anyone can host Project Aura - no one needs to use the hosted version to take advantage of Project Aura. However, if you do use the hosted model you get the advantages of not having to build out the infrastucture yourself and you can take advantage of a much larger dataset since all recommendations from the hosted service can take advantage of the taste data from all of the participants.
Post a Comment