tag:blogger.com,1999:blog-7434008.post7234436736433292455..comments2023-11-02T09:04:29.998-07:00Comments on Adrian Cockcroft's Blog: Hiring a researcher at NetflixAdrian Cockcrofthttp://www.blogger.com/profile/14695336135416848505noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-7434008.post-10676811348405006752008-03-01T01:10:00.000-08:002008-03-01T01:10:00.000-08:00Hi Vern, this was suggested and discussed a while ...Hi Vern, this was suggested and discussed a while ago on the Netflix blog. I agree that more resolution in the input should give better results, all else being equal. However, all else is not equal and Netflix did test half-stars a while ago, and found that it adversely affects the way people enter ratings, and we get better results over-all when users can only enter whole stars. This is a good example of the way Netflix does development. Intuition is used to come up with a plausible improvement, but measurements are used to refine, confirm or reject that intuition.Adrian Cockcrofthttps://www.blogger.com/profile/14695336135416848505noreply@blogger.comtag:blogger.com,1999:blog-7434008.post-72228134588277100682008-02-29T15:34:00.000-08:002008-02-29T15:34:00.000-08:00Adrian, the 5 point rating scale may be a limitati...Adrian, the 5 point rating scale may be a limitation to improvement in an algorithm predicting customer ratings. <BR/><BR/>Since Netflix customers rarely rent, and rate, films they don't like, the 5-point scale is functionally a 3-point scale (3, 4, and 5). <BR/><BR/>The 5-point scale, with its 3 points that are used most of the time, and the other 2 that are rarely used, could be converted, without losing existing ratings data, by adding the values 1.5, 2.5, 3.5, and 4.5. <BR/><BR/>Then Netflix would have a 9-point scale, of which 7 values would be frequently used, and thus allow quantitative models to predict with greater accuracy than now.Vernhttps://www.blogger.com/profile/14212634852140174098noreply@blogger.com