Netflix Movie Recommendations

Karl writes:

My recommendations don't work very well for me. I've rated 2,000 (two thousand) movies as of today and that should be a sample big enough to make their algorithm work, I think.

The real problem seems to be that a single instance of a high rating will spawn a lot of only vaguely similar movies -- I liked Finding Nemo and was deluged with Disney, I also found my science fiction queue thereafter flooded with anime...

Netflix uses a proprietary movie recommendation program called Cinematch to evaluate your ratings and recommend movies. How good are the recommendations you get from Netflix?

Comments

[prodly]

Same here. I've rated like 1000 movies and they got nothin for me.

[Ed Neely]

I have also found that the last rating I gave is weighted too heavily. I gave "Grapes of Wraith" a high rating, then got nothing but old Westerns until I rated a new release. Gave a good rating to a TV series DVD, and got a bunch of TV DVDs. So, it's not perfect, by a long shot. But, every now and then a movie I've always wanted to see, but never have, pops up and I add it to my list. So the function is not worthless. Just hit or miss at times.

[Aron]

I find it disappointing as well for my personal use. The company frequently suggests that a very significant percentage of rentals comes from it (perhaps 70%, can't recall exactly). Mainstream renters and power users most likely have different patterns.

I think a bleeding edge search mechanism would function better for me. I tend to watch movies in sequence that have a common thread (same Director, same actor, similar theme, timeframe, genre, award winners, highly rated, suggested by some person X, etc). I'd like to be able to navigate rapidly (top notch web code) through the potential base of movies in such a way.

If they want some blog PR from riding the buzz wave they could implement a tagging process (or auto-strip from user comments and 'my 2 cents') and let the tags be potential axes for search.

[brian d foy]

The Netflix Recommendations are pretty awful for me, even after I had rated over 1,200 movies. It seems to me that they have more to do with the category of the movie than anything else: if I say that I like a western, such as Forgiven, I get a bunch of westerns in the recommendation. I really don't like those, in general. If I rate a couple of Cary Grant movies highly, I'm deluged with recommendations for classics. It's been the same for foreign films too, which somehow

The kicker of the whole thing is that they keep recommending me the same thing every time I'm there. I have a new rating strategy though: rate towards the extremes. Instead of rating things 3-stars, which I mostly do, I'm using the 2- and 4-star ratings more. If I didn't like something, instead of using a 2-star, I give it a 1-star. All TV shows get a 1-star even if I like them: I can TiVo those, so I don't want to see them in my NetFlix ratings. I think this strategy has worked a little, but I still am not interested in most of the recommendations.

[manuel]

I've found Netflix recommendations not helpful. I've just compared my personal recommendations from Netflix, GreenCine, and Blockbuster:

Netflix:
1. All Quiet on the Western Front
2. Stand By Me
Anime Recommendations
1. Dragon Ball Z: Imperfect Cell Saga
2. Inu-Yasha: Affections Touching Across Time

GreenCine:

1. Ping Pong Club vol. 5 Rots In Hell
2. Crouching Tiger, Hidden Dragon
3. Vampire Hunter D
4. Akira
5. Cowboy Bebop: Session 5

BlockBuster:

1. Cujo
2. Winter Kills
3. Bring It On (2000)
4. Hell of the Living Dead (1980)
5. Heavens Above (1963)

Although I've rented almost exclusively anime titles with a few Japanese and Korean motion pictures, Netflix top two recommendations are neither. Under the anime section Netflix recommends Dragon Ball Z which is oh so typical. As the poster above mentions, Netflix keeps recommending Dragon Ball Z which is becoming tiresome. Netflix did surprise me with Inu-Yasha as this is one of my favorite titles. GreenCine did a better job picking recommendations: All except one are anime I would be interested in. Crouching Tiger, Hidden Dragon is a martial arts flick a la Matrix which looks interesting. Blockbuster was completely off mark with their recommendations. I've rented (and rated) anime exclusively from them yet none of their recommendations include anime.

[Tony Gentile]

The only time NFLX recommendations work for me is when it recommends something that I've seen before, but haven't rated in their system (i.e., one of the hundreds/thousands of movies I rented from Blockbuster in the decade before I started using NFLX).

For me, the main problem is that I can't manipulate the recommendation algo with direct preferences, e.g., "Don't recommend any movie over 10 years old".

I get far more interesting recommendations from my **Friends** list... I'm disappointed that no one else has mentioned this (yet), as I wonder what that means for adoption of the feature...

[Pinball Mars]

I actually think Netflix's recommendation system is alright. I particularly like how some DVDs have TWO number ratings assigned to them. A rating from the general Netflix base and an average rating from "people who rate like you".

I'm in a situation where the Netflix recommendations work for me for, in my opinion, two reasons:

1. I rent a pretty eclectic batch of movies. My big interests are cult movies, sleazy b-horror movies, classic films, and art films. Netflix's recommendations (particularly for classic movies) have been pretty good about informing me of things being ON DVD that I didn't even know had been released on the format yet.

2. I'm one of those users whose queue is always at the 500 disc maximum. The recommendations system at Netflix (the shooting stars, the average of people who rate like me), more than anything else maybe, sometimes help me decided which movies to move up to the top of my queue, depending on my mood.

When rating movies, I have the same method as the guy above who generally rates in extremes. I tend to give old cheesy b-movies from the 50s and 60s 5-star ratings (even ones I don't like that much, sometimes) because I like those movies and I'd like the NF recommendations system to recommend MORE old b-movies to me. Stuff I forget I wanted to see, stuff I didn't know was on DVD or didn't know that Netflix had. It often works.

[Anand]

Personally, I've found some movies I've really liked from Netflix's recommendations system. There is definately alot of crap there too though. It seems to think I'm Woody Allen's biggest fan even though I've rated the few of his movie's I've tried pretty poorly.

I wish it'd some more options as well. The ability to restrict it to movies within the past X years for example. It'd also be nice if it could filter out all of the TV show DVDs.

My Netflix thinks I'm gay!

[Amy]

I have added many, many of their recommendations to my queue - and have also rated a thousand movies. But, I am devising rating software to remedy the rating problem. Give me six to nine months to complete it. But, I think it will really help all of us movie fanatics!

I rented Shark Tale and now my recommendations are all for Anime!

[Aron]

Amy, if you have not already, I recommend reviewing: http://movielens.umn.edu/main

[roror]

I'd be interested in your software Amy. Just curious about what method you are using for your program. Some collaberative filtering approach or something completely different?

I'm glad I read this thread. I am fairly new to Netflix & have rated 135 movies, and the recommendations are worthless. I see more experienced people haven't had much luck with them, so I won't bother rating movies anymore. (There are good ways to find good movies: compilations of "ten best" lists, reviews by reviewers who seem to like the same things I do, etc.)

[Kate]

Hollywood Video has recently added recommendations to their website and I've found them to be much better than Netflix's weak recommendations. Hollywood recs seem to be more on target to the kinds of films I like and so far has been a good source of queue ideas. I use their site in tandem with my Netflix account.

[James Andrew]

I've rated almost 900 movies at present, and I too have had horrible recommendations. I've contacted customer service about this a number of times. I believe the problem lies in the rating system. Netflix only lets you rate movies on a 5 star rating system. To me it breaks down like this. 1, is a terrible movie. 2, is a movie that is flawed or that I didn't like but it has a few slight redeeming qualities. 3, a movie that was good. Nothing incredibly original, but something that held my interest and was worth watching. Not a movie I'd want to own. 4, A great film. Not perfect, but one that I'd love to own on DVD. And 5, a perfect film. As I rate movies based on this scale, most of my ratings end up being 3 stars. I prefer Blockbuster, they allow you to rate movies with half stars. So your ability to rate the movie properly is greatly increased. And even better, I love the YahooMovies rating system which allows you to place a letter grade (i.e. A-, C, D+) on the movie as well as having a comprehensive description of exactly what your rating means. I think if they could expand their rating system to include more options that recommendations would improve greatly. As it is, it's too hard to clearly identify which movies you really love.

[Aron]

I have been watching the Sopranos, within which they watch the ol Cagney film 'Public Enemy'. Surprisingly, my recommendations page just pulled up that movie which is a good hit. But under 'why is this film recommended?' there is no indication of the Sopranos as the cause.

Coincidence perhaps. That kind of knowledge in the system would be useful, especially if the connection can be explained on the website 'this movie is one of Tony's favorites in Season 3 of the Sopranos'..

I get the impression that Netflix has cut off significant funding for development of cinematch. If so, that's not a good move. The Cinematch technology fulfills an important role of extending subscribtion lengths, and reducing emphasis on new releases. It is also something that will continue to exist in a VOD world. With the proper system in place, the Netflix subscriber base might provide the data entry labor for detailed connections between movies.

[Aron]

I think cinematch is locked into one of those local\global maxima problems. It is being tweaked perhaps to perform better but within a very limited scope. They need a 'Netflix labs' to throw out some potential alternatives who's basic design is significantly different. ;)

[REN]

I always have trouble rating movies. It just feels wrong to me to rate a good martial arts event with 4 stars and then give the same rating to a classic movie. I realize when considering the rating you take the genre into context and the software likewise does the same, and recommendations given I hope aren't based solely on ratings.

I think the ratings should be expanded to include additional categories. It would still be a quick process to vote, just rate the movie in maybe 5 categories (5 clicks). Some ideas for categories would be "Overall rating", "Genre rating" (i.e. very good as a low budget B sci-fi film), "Picture Quality", "Extras", etc.

[M-Class]

I stopped using ratings even though I thought they really were giving valid recommendations. The problem I had is that on three seperate occasions about a third of my ratings have vanished from the NF database. As a result I keep getting recommendations for stuff I've already rated and/or watched with Netflix! I gave up fixing my ratings when this happened for the third time. A broken feature is useless to me.

[Skro]

Netflix gave me terrible recommendations even after I rated over 1000 movies. Then I went back and removed the rating on all the random 3 star movies and only rated things I loved or hated. Everything else gets a no opinion/not interested. It seems to work much better now.

[Sonja]

I don't even bother using the recommendations. I pretty much know what I want to watch DVD wise, so I don't use the recommendations to find new movies to watch. This past weekend I watched Shrek 2 and Shark Tale and got some recommendations that I just chuckled at.

[Peter]

One of the wonderful opportunities of a rental system like Netflix is the opportunity to provide feedback on whether you liked a movie or not. With modern technology, one would think that a competent recommendation system could find a few "diamonds in the rough" that I would have otherwise missed. I have rated over 1600 films for Netflix and their recommendations for me are still mediocre. The Netflix recommendation system is especially bad on comedy. It regularly recommends comedies for me that I just hate. I know that this poor performance is not necessary since Hollywood Video has a system that seems to be much better. Why can't Netflix, with all the recommendation data it has, do as well as Hollywood Video?

[jeff alin]

I such dick

bitches

I have 3 queues; one for me, one for my son & one for the rest of the family. My son rents anime under his profile but under my profile ALL I get are anime recommendations! I have never rented anime under my profile!

[lcd_cow]

I have had a lot more success with ITOG (http://www.itog.com). It's a relatively new site, and they do a lot more than just movie recommendations (also books, music, and a bunch of others), but their algorithm has definitely been working MUCH better for me than Netflix's has.

Check it out.