DanC has moved his family out to Ames, IA, one time home of Neal Stephenson. And look, the neighbors had a welcoming party!

Good luck at your new home, Dan, Heidi and Anna


Pigs and Features

Both of these efforts add a pretty significant twist to the basic premise of rating movies and getting predictions. The first, MoviePig, doesn't actually ask for specific rankings. Instead, the user is asked indicate preferences between movies, building a ranked list as they go along. In this, it seems to exist as the extreme version of the old Movie Critic site's "Sanity Check." But, because the ranked list isn't tied to any labeled values ("One-Star" / "Hated It"), you end up creating complexity in your ratings that you've never seen before. Fortunately, you can cluster movies together, though this isn't intuitively obvious. MoviePig also compresses your ranked list to fit on screen, though you can force it to uncompress a particular section.

You can ask MoviePig for a prediction on any film in its database, and generate recommendation lists. Topping the list of current releases for me is Eternal Sunshine of the Spotless Mind (a match with MovieLens), which it places between the clusters marked by School of Rock and Intolerable Cruelty.

MoviePig's interface is drag-and-droporific. Everything is drag-and-drop, including picking a film for prediction (drop it on the pig), and everything is done through Flash. The ranked list gets confusing if you try to reflect any subtlety in your selections, which is encouraged. All this makes it easy to get started, but you can start ranking and getting predictions before you create a userid, which may add to the confusion. Between the so-easy-it's-difficult interface, and the fact that MoviePig's movie database only contains movies released from 2000 and later, I can't say that I expect to make regular use of their system.

By attempting to make entering of opinions extremely easy, MoviePig is addressing one of the main criticisms of collaborative recommenders. Mainly, that entering ratings for a bunch of movies is a pain. Recommendz moves in the direction of making the entry of opinions even more difficult.

Recommendz requests input from the user on a per-movie basis. For each movie, the user enters an "overall rating," as well as the option to select from a list of 40-plus "features" that can be associated with the movie. After selecting the feature, the user identifies the quantity of the feature in the movie, i.e. "There is 0/10 of Adam Sandler in Moulin Rouge" or "There is 8/10 of music in Moulin Rouge" and then the user is expected to rate on a negative-to-positive (neutral allowed) scale how the feel about the quantity of the feature in the movie.

As you can guess, many of the features are very inappropriate. The system allows users to suggest features that don't already exist for any movie. But for this information to be valuable, data needs to be associated with one of the other movies. In this sense, you are giving Recommendz a more nuanced picture about why you feel a certain way about a specific movie.

Recommendz is an academic project based at the Mobile Robotics Lab at McGill University. The premise is that this detailed information will allow the system to generate better collaborative recommendations. And I have to say that I am not displeased by the recommendations that I have received from them, but data entry is a big headache. There is no way that you can do bulk data entry to get started. It requires a minimum investment of twenty-minutes to figure out the interface and rate enough films and features (including rating some movies you hate) to start getting results.

UPDATE: I have added these two recommenders to my sidebar. I mentioned this weblog as a possible resource to my Knowledge Discovery professor, so it seemed best to add them to the list.

Page 23

This seems to be another one of those weird things going around. Maybe it'll get some of my juices flowing.

"The number of permutations of the four letters a, b, c, and d is clearly 4! = 24."
-- The fifth sentence on page 23 of Probability and Statistical Inference, 2nd Edition by Robert V. Hogg and Elliot A. Tanis

Hogg and Tanis, sitting to my right, seems to be closest by a matter of centimeters, just barely beating out Neil Gaiman's American Gods on my left.

I picked this "meme" up from one of the many Microsoft Blogs collectively syndicated at Weblogs @ ASP.NET, where it seems to be doing the rounds.

The meme says:
Grab the nearest book.
Open the book to page 23.
Find the fifth sentence.
Post the text of the sentence in your journal along with these instructions