Comments on: Screwy Netflix Recommendations

By: Karyla

Karyla — Tue, 29 Apr 2008 00:54:17 +0000

I have had pretty good luck with my Netflix queue. But, I totally agree with Kieran–you have to have a certain amount of skepticism when viewing the recommendations. I rate movies that I think are really horrible, too, especially ones that Netflix recommends that I’ve already seen. Are you doing that?

By: Kieran

Kieran — Mon, 28 Apr 2008 18:51:37 +0000

DR, your math may work out well, but that still doesn’t get at the human aspect of enjoying a film (or detesting it, for that matter). What is at issue here is Netflix’s claim that they can accurately quantify the quality of a movie as it compares to customers’ tastes. That’s a noble and bold claim.

Suffice to say I have stumbled across some really interesting plot descriptions on the Netflix recommendations, but have yet to say that it’s pointed out a hidden gem to me. I can usually tell which elements of the recommended film brought it up on my profile’s radar, but there’s more to a good movie than a algorithmic string of similar elements.

I guess I approach the recommendations with a healthy dose of “Caveat Emptor.” I know when I am watching a recommendation from Netflix that I may send it back without finishing it to hurry along that next movie in my queue.

Kieran

By: DR

DR — Mon, 28 Apr 2008 15:04:05 +0000

With 900 ratings you should be getting recommendations with AVERAGE dedviation of less than one star from your actual rating of the recommended movies.
Why don’t you try the following test: watch any 10 movies but DON’T rate them on Netflix yet, but do write down your rating for each one of them on a paper.
Now – go to Netflix, run a search for these movies and calculate the average difference between what you rated and what Netflix engine thinks you would rate them. You will be surprised to see that I was right.

Brian – the recommender engine does not work by actor or genre. It is not as simple as that. If you put some effort into studying the math behind the recommending algorithms, you may change your mind about their capabilities and validity.

DR
(A Netflix-Prize competitor)

By: Brian

Brian — Fri, 25 Apr 2008 23:47:17 +0000

It’s impossible to make accurate predictions on what people like. For instance: How can it know that I won’t be able to stand watching “XXX” Staring Vin Diesel (mid 90’s)based on me giving the original “Punisher” (late 80’s) a four star rating?

It simply doesn’t know good-crappy movies from crappy crappy movies.

On a similar note, I cannot stand a large majority of independent films but there are a few I really like so they get high ratings. It sees those high ratings on a few indie movies and baaam! Netflix thinks I like all crappy indie movies.

It can be confusing based on actors too… say you like John Travolta: Okay, you get Battlefield Earth.

What about Robert De Niro? Okay, we’ll suggest Shark Tale.

Moral of the story, robots can’t think.