CinefileNotes

From MWCSWiki

Why stroke play makes more sense than stroke percent: Because even if the sizes of the ex/cex are radically different, they are radically different no matter what keyword we're considering. And because stroke play is better than stroke percent given the choice, because stroke percent is going to factor in a keyword's absence too much.

Why match percent makes more sense than match play: Because you can't do match play intelligently, the bigger category is going to hopelessly outsize the smaller.

---

--Stephen 15:50, 26 January 2009 (UTC) Here are our candidate raters:

• Naive Bayes.
• Stroke play.
• Match percent.
• Wednesday, without poison.
• Wednesday, with poison.

Only one TLG. It asks the rater for the keywords that it (the rater) is most interested in knowing about. (How many of these? Dunno. Does the rater sometimes return null or throw an exception to indicate, "actually, I'm good now" ?)

The TLG then generates two sublists. One consists of movies that contain the keywords the rater requested. The other consists of PURELY RANDOMLY CHOSEN FILMS. Info from the first set of feedback is fed back to the rater so it can improve on itself, and this leads to iteration. Info from the second set is used to calibrate the category likelihood threshold.