Collaborative Filtering
The CALLERLAB Women in Calling committee just released the results of its member poll on favorite patter records. The records selected are listed at Supreme Audio (here’s a URL that might work: Women In Callng, but I wouldn’t count on it; Bill’s site does complicated things and assigns a session number when you first enter. I’m giving you the URL without the “shopping cart id”.
Anyway, the recordings are listed, but without the actual tallies (most records got one vote, but a few got a couple, and Kayla Rae/Jake (5) and Braveheart (4) got several. Too bad Kayla Rae/Jake is out of print; it’s a real classic.
I took part in the survey, but I noted to Deborah (Deborah Carroll, chairperson of the committee) that my choices change frequently, based on time (I get new favorites all the time), audience (I usually don’t do screaming electric guitars for the dancers at the Palo Duro Senior Center), and purpose (workshop? weekly dance? special dance?). So I’m not sure how useful my selections are.
However, the list got me thinking about collaborative filtering. There were many selections on the list that I use and like, and others that I really don’t like. If I knew who picked the records that I like, I could then look at other choices of that caller and maybe I would like them also. On an anonymous, large-scale basis, these are known as collaborative filtering recommender systems. You can see this at work at Amazon; as you look at books, Amazon will show you other books that people who like the current book have also liked.
What if, as we selected records at Supreme, the system would recommend other records based on our past purchasing history and the purchasing history of others who have purchased similar collections of records. Of course, merely buying a record doesn’t guarantee that I’ll like it once I hear the whole thing, so it would be even better if there were a record-ranking system. After I’ve ranked some records, the system could recommend other records that I might like. To see a non-commercial system at work, check out Jester: The On-Line Joke Recommender. Rate a few jokes and Jester will recommend other jokes. As you rate more, Jester’s recommendations become more precise.
Now that would be cool!
