101 comments

[ 2.1 ms ] story [ 162 ms ] thread
Is this supposed to be a Show HN?

Where does it get your move list from? I know I rated a lot of movies on my rotten-tomatoes profile, can this be used?

EDIT: Also, it's not clear that you can start using the product without Facebook, I like the fact that it lets you rate a couple movies before asking you to sign-up (with Facebook or with a regular email/password)

Facebook auth is also failing for me as I receive an internal server error during auth.
Hey! I work on the site, sorry for the mishap. You should be able to authenticate now!
Would be awesome to have import option from CSV file. After jinni.com folded, I've been tracking my movie ratings in spreadsheet file.
Did this work for anyone? Meaning; anyone got good (for them) suggestions? Where does the data set come from?

Am I blind or did I miss the button for 'I did not see this movie'? Because I didn't see quite a lot of the ones that I was asked my opinion on.

All in all, I always like these kind of things, but asking me if I like that Star Wars drivel 3 times in a row and then asking me if I liked 4 movies I never saw skewed my results a bit I think.

Keep improving though! If I can get only one 'wow' movie I never saw recommended, I would be really happy.

There's a "Haven't Seen" button right below the 4 circles. (At least for me :/)
A there is not for me (Chrome on Linux). I found it by trying some zooming options. Thanks.
I'm impressed. It gave me what I thought to be a small, weak data set, but the recommendations were better than those I get from Netflix, which has much, much more of my data to work with.
I was surprised too. I didn't really connect to any of the choices during building the profile but the recommendations were right on.
Same here. It showed me a bunch of blockbusters that I either haven't seen, were just kinda good, or even meh.

And came up with amazing recommendations of really cool movies I've either watched and loved, or haven't seen but always wanted to. Even the top pick for "movies from this year" was the one movie where I saw a trailer and said WANT.

What makes the rankings better than you get from Netflix? I'm impressed that it managed to suggest Clue to me, but I haven't seen the rest, so can't really say if I would like them. (More, if it just suggests movies I've already seen and liked... hard to say it did a good job, or that I'm just predictable in cliques of movies. :) )
The point of this site is to give you recommendations you haven't seen. I hate it when all the recommended movies on Netflix are all movies I've seen and can't get rid off...
These recommendations are pretty crazy (in a good way). All my favorite sci-fi, comedies and thrillers in one list. Impressed. Going to check all the ones I haven't seen
Agreed. I just don't know how I could "rate" that the site is recommending good movies until I have watched them.

That is, I'm curious how so many folks are claiming it is doing a great job.

They all look interesting. Many years ago, netflix did the same. It recommended me a bunch of movies that seemed really interesting and I liked a ton of them, but these days netflix is a low bar. I'd bet they could do a great job, but have a more limited movie selection. Either that or the algo has actually declined.
Or payolla. Never assume a bad algorithm when a good algorithm makes less money.
Are you just on streaming? I've noted that the recommendations for streaming are fairly restrictive since they only have around 6000 movies available at a time.

With DVDs, the greater selection means better recommendations.

    > they only have around 6000 movies available at a time
That many? Always feels more like 10.
> the recommendations were better than those I get from Netflix

I can tell you exactly why Netflix makes such poor recommendations, and why almost anyone can do better with modest effort:

Netflix has to give recommendations for you from the 6000 movies that it's currently showing[1]. They can't recommend movies that they don't have. Whereas Taste.io can choose from the entire universe of ~500,000 movies.

An example should make this clear: If you liked The Godfather, it's an easy prediction that you'll like The Godfather: Part II and Part III. Suppose Netflix is currently showing The Godfather, but not the sequels. They cannot recommend the sequels to you. But Taste.io is not bound by that restriction; they can in theory recommend any movie that exists. It's much easier to find matches among 500,000 movies than among 6000.

[1] Netflix has just 6332 movies in the USA as of this date and even less in other countries (eg., 4365 in Canada). Most people are surprised by how few movies Netflix actually has. The Netflix user interface makes it very difficult to get a good impression of the number of movies; you can't just scroll alphabetically through the entire list for example. Source: http://netflixcanadavsusa.blogspot.ca/

Presumably Netflix are using a wider system to choose which movies to licence? I wonder what their process for adding movies is - do they have a list from each studio they work with and select a movie to add, out do the negotiated each one separately.

Seems they could have a not yet available category that would let people pre-order; they could recommend a far wider swathe of content then.

Also why the Netflix reccomendation system used to be a bigger selling point for them before they switched to streaming.
> An example should make this clear: If you liked The Godfather, it's an easy prediction that you'll like The Godfather: Part II and Part III.

Really? I've always heard such negative things about Part III that I never bothered to watch it.

I get what you're saying, but Netflix seems even worse than that. To wit: taste.io recommended movies for me - which I enjoy and that are in Netflix's current catalog - that Netflix hasn't recommended for me.
Cool user flow. I bet you could hit me with another facebook login request with the "see your results" popup, and I might not even mind it. That said, I won't click it.
I am curious why you used the user-closeness model.

Matrix factorisation won the Netflix prize: http://dl.acm.org/citation.cfm?id=1608614 . What made you go for user closeness?

We've tested other models but collaborative filtering gave us the most "human"/natural results. Also, didn't Netflix toss the matrix factorization after the contest? IIRC they decided to keep their current algorithm.
Interestingly, my sense has been that Netflix's recommendation engine does a poorer job now than it did 10 years ago. I always assumed that it was because they used to use fairly straightforward collaborative filtering, and now they seem to be heavily focused on looking for stuff that's somehow cosmetically similar to other stuff I've watched.

So, like, instead of saying, "You liked Nosferatu? Well, other people who liked Nosferatu also liked Ran, so let's suggest that," it now goes, "Hey, that's a vampire movie! How about Blade?"

I love the fact that the ratings are semantic, and limited to four easy to understand values rather than 5 stars:

    Awful (Can I have those two hours back?)

    Meh (Not great, but better than nothing 
         to kill time, escape or veg out)

    Good (I enjoyed watching it)

    Amazing (I'd watch it again and recommend
             it to friends without hesitation)
With 5 stars, everyone interprets 2, 3 and 4 stars differently, e.g.:

    Horrible  Bad  Meh   Good       Best

    Bad       Meh  Good  Very-Good  Faves
Even the same person over time will not use a 5-star scale consistently. Even when I try to be consistent (I use the latter values for Netflix), if I like a movie but don't love it I don't know whether to give it 3 or 4 stars. On different days in different moods I'll make different choices.

I've no doubt that unreliable ranking data made Netflix recommendations harder, and impacted their mix of recommendation algorithms -- i.e. leaning more heavily on those that work despite rating scale inconsistencies. I'd expect the mix that works best for Taste.io will be different.

I have to agree that the rating options are very refreshing and easier to keep consistency over time. It also forces a choice between positive and negative which I'd imagine helps the algorithm learn things faster. Hmm...
what database (movies API) are you hitting ?
Currently using iTunes but will be moving to TMDB soon
I already maintain a list of movies to watch based on whatever interests me on netflix + random lists on internet. I took their quiz and their recommendations had good overlap with my existing list. Impressive!
Are others getting a lot of recommendations for old movies? I have to be in just the right mood for a movie from the 1950s or earlier...
You can use the filter on the browse page to filter out old movies, or simply hide them!
Very interesting. This did two things:

1. Correctly identified a bunch of movies I had seen and really liked + some promising ones I hadn't seen, and 2. Showed me that I don't like any new movies

None of the 2016 movies were over 70% for me, whereas it identified some movies I had seen and loved as 97+%.

How does this work?

I had a similar experience. Every highly rated recommendation was either one of my favorite movies or on my list of movies I need to see very soon but haven't made the time for yet.

It knew to put The Holy Mountain at 100% based off me loving the new Mad Max, feeling ambivalent about the Star Wars prequels, hating the 2007 Tranformers movie, and thinking Anchor Man was good but not great. Like, how the hell? It's spot on, but how did it get that from my input? Is there more info on how this thing is getting its results? I would love to see even a sketchy outline of the algorithm.

Very cool. My list of recommendations included movies I greatly liked, as well as ones that have for a long time been on my to-watch list.
Suggestion of 'Spirited Away' based purely on my ratings of X-Men and Saw? Spot on! I would love to read about the algorithm you guys are using (in an upcoming paper perhaps? :))
Indeed! I'm baffled. Is there a dataset of film style this is drawing on, or is it purely based on prior user trends?
Very impressive.

Recommendation are excellent - I noticed that recommendations are NOT about movie genre but about how the movies are made, their point, story, etc.

I.e., I hate when Netflix thinks that I like all stupid vampire movies because I liked Thirst [1].

[1] http://www.imdb.com/title/tt0762073/

Nice! I'd love to be able to point it to my IMDB data. Both because it's got a comprehensive list of my ratings, but also so that it knows which movies I've seen to avoid recommending them!
Has this been around for a while, and gotten a recent facelift? I remember something very similar (and may have even been called Taste) back in 2013.
The site is only a few months old, if you could remember the other site I'd love to know.
Is there any way to filter the recommendations down to the offerings from various content providers?

E.g. 'Available streaming on Netflix'

Coming soon!
Awesome, thanks for this.
I've rated many films on Netflix. It would be cool to be able to import my ratings from Netflix (or similar services like Amazon Video or IMDb) into Taste.io. Your service would get a lot more data to work with. :)

So far the recommendations have been very good and I've bookmarked a couple films to watch later.

That suggested some obscure movies that I like, wow!
Playing around with it I'm finding the interface notably more friendly than MovieLens.org for inputting ratings, but the results seem worse.
Can you please show release dates? Sometimes I'm not sure if a movie is the original or a remake.

EDIT: The release date is included in the URL for the movie.

I've love to see this integrated with a TV/Movie scrobbler such as trakt.tv.
Nice, the rating consensus system makes me curious about some stats, like for all users, the best movies with high consensus, and a list of good movies with least consensus.
Interesting question, just ran a quick analysis for all movies rated 4.5+ star

Highest Consensus:

1. Whiplash

2. 12 Angry Men

3. Se7en

Most Controversial:

1. Birdman

2. Citizen Kane

3. 2001: A Space Odyssey

> 1. Birdman

I knew that abomination was polarizing, but i did not quite expect that.

Abomination? That was oone of my favorite movies of all time!
;)

It's an emotion vs intellect thing. Birdman is like cocaine to people who like the emotional side, and like hydrochloric acid for people who want a movie to be rational.

I'm just surprise the numbers even bear that out like that.

There must be problem with your dataset then. Not enough people? Everyone in this field knows that Napoleon Dynamite is the most controversial, ever.