- 11comments
- 200comments
- 7comments
- 62comments
- 298comments
- 148comments
- 128comments
- 14comments
- 100comments
- 66comments
- 42comments
- 45comments
- 26comments
- 26comments
- 34comments
- 7comments
- 4comments
- 6comments
- 21comments
- 246comments
- 49comments
- 1comments
- 57comments
- 112comments
- 46comments
- 23comments
- 35comments
- 22comments
- 164comments
- 47comments
The posting is worth reading but the premise of bias is wrong. Technology (that is, Netflix's recommendation algorithm)does not have a racial bias per se but it does discover the cohorts viewers belong to (using their behavior) and report categories of recommendations derived from the aggregate cohorts behavior. If race were not a distinguishing feature in the cohort's data, it would not be a distinguishing feature.
Recommender system algorithms typically personalize search results based on similarities to items recently watched, purchased or simply added to a shopping basket. More advanced recommenders suggest based on past history (well YouTube seems to). I don't find personalization by similarity offensive. If I watched a Netflix movie starring Sidney Poitier, I'd expect recommendations to view movies from top Black actors of all time. I see this as ranking on similarity not on racial bias. http://www.imdb.com/list/ls051979184/
Exactly right, but that wouldn't result in as controversial a headline.
Yup. When I was reading the Horatio Hornblower books a while back, and Amazon suddenly started recommending books by Patrick O'Brian and Dudley Pope, I didn't arrive at the conclusion that Amazon's algorithm was "racist" with respect to 19th Century British naval culture.
Title should say the algorithm exposes an average viewers preferences. The algorithm is making recommendations based upon user choices and isn't biased. Not really a fair characterization.
Yes and no. The algorithm works upon manually-defined tags, and one of the points being made is that those are too broad, possibly because the people tagging don't have enough insight into particular populations.
The example given of the person who got recommendations for films with lesbian sex scenes after watching Blue is the Warmest Color is relevant. Yes, the film has such scenes. But it's also won the Palme d'Or, yet that isn't tagged and therefore used by the algorithm, when (I'd say) it's much more likely to have been the reason for a particular person to watch it.
Blacks are about 20% of the US population, while Netflix has some demographic information I can't really imagine they are tracking race of the viewer (I could just imagine the shit storm that could erupt if they did), they do however classify films based on various categories.
Netflix didn't invent black cinema people did and way before Netflix was even an idea in some one's head here are for example Black Cinema film festivals:
American Black Film Festival Freaknik Hollywood Black Film Festival International Black Women's Film Festival The Kidflix Film Fest of Bed-Stuy North Carolina Black Film Festival Pan African Film Festival Roxbury Film Festival San Diego Black Film Festival Women of Color Film Festival
The US also has "black only" film awards including the African-American Film Critics Association Awards, Black Reel Awards, Black Movie Awards and so on.
So after watching multiple movies that most likely their only common denominator was "Black Female Directors" Netflix suggested you a category that contain additional movies that fit your recent viewing habits.
On a side note, I watched the new Crouching Tiger, Hidden Dragon and the Marco Polo special on Netflix recently and all of a sudden what would you know? Chinese / Asian martial arts films appeared and took over my suggestion feed.
However after binge watching Cosmos and The Inexplicable Universe I wasn't greeted with the suggestion of the category "Documentaries by Black Astro-Physicists" but rather with "Space and Nature Documentations".
As of the last census, blacks were 12.2% of the population, not 20%.
https://en.wikipedia.org/wiki/Demography_of_the_United_State...
So Netflix is biased because it doesn't feature African-American movies as a default category? What about Filipino movies? Should Netflix turn their default recommendations into a progressive stack?
The article makes a few interesting points, particularly regarding the difficulty in choosing the categories/tags with which to classify things, and also regarding the dangers of filter bubbles in recommendation algorithms.
Unfortunately, it seems the conversation here is completely focused on defending Netflix from a perceived accusation of racism.