justaddwater.dk

[Jesper’s note: It will be interesting to see how this maps to Jakob Nielsen’s old ideas about reputation managers.]

Netflix example:

Wall-Mart went head to head with Netflix and lost. They are now linking to Netflix

* 4 million subscribers 10 months ago
* 60,000 titles, 40,000 of which are rented every day
* 2/3 of movies rented come from recommendations

The movies we recommend generate more satisfaction than (new releases)

– Neil Hunt, Netflix chief product officer

Collaborative filtering

Example with 5 users
1. Look for people with similar habits
2. Choose the best fiter. In this case, user 3
3. Find differences in best filter
4. Those difference become best filter

Netflix recommendation engine contest
* Giving a huuuge user database
* Improve recommendation by 10%
* in one week a team came up with a suggestion that was better

Implicit or Explicit behavior

Implicit data: Not a rating.
Example: LastFM where user don’t rate, but just watching what what music you play.

Implicit is everything behavioral data.

Explicit behaviour is when you demonstrate preference

Pandora’s Music recommendations
Hired musical experts and recorded small facets of the music.
Not a personalized recommendation system. Music based system recommendation from musical elements

Tivo (movie recommendation) thinks I’m Gay
* Basil Iwanyk
* he started watching “guy stuff” like war movies
* tivo started recommending documentaries on The Third Reich
* Queer eye for the straight guy
* Later this became an episode of King of Queens

So, recommendation engines are not perfect.

Customers who bought (clothes) also wear:
* clean underwear
==What?!?

Changed to
Customers who wear clothes also shop for:
* clean underwear

Wall-Mart put in a recommendation system last year
* planet of the apes
* recommended a martin luther king video

Still have not put recommendations back on site.

[My live notes from UI Conf. All UI Conf notes. Expect updates of thes rough notes, with references, images, etc. within the first few days – this footnote will stay, though. Feel free to add comments, links to similar notes, presentations, correct quotes and add where appropriate. thanks! Also I post my pictures on Flickr tagged uiconf]

Technorati Tags: joshua porter, uiconf, ui11, recommendation, reputation, jakob nielsen, uie, bokardo, filtering, social, netflix, niel hunt

This entry was posted by Jesper Rønn-Jensen on Thursday, October 12th, 2006 at 22:16 (GMT-1) and is filed under Events/seminars, justaddwater.dk, ui11. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.