Pigeonholing is the reason I can't stand Pandora radio. I like novelty, and I want someone to recommend types of music I haven't heard before. Pandora's model takes what you already like and gives you more of it. How is that useful to anyone but the closeminded?
Actually, this isn't quite correct. At least, not for a station I've been listening to for the last few years (3-4). It started out based off of OK Go, and has continuously introduced me to new music that I either like or dislike (it does keep trying to get me to listen to nordic death metal). I've learned about a lot of new bands, and have actually managed to develop a (much more accurate than my previous) sense of "current" music.
Maybe Pandora's model changed, but the one thing that set it apart from other platforms is that its music recommendation is not based on 'other music that people like who also like this track/artist', but rather on a number of properties of the track itself.
As a result, you get recommendations that are musically similar, but often new and unexpected.
The indexing is based exclusively on mood and the ratings are user-contributed. I can forward contacts if someone wants to know more about the project.
How many ? As far as I know, as many ratings as there are movies. Of course, the more ratings the more "accurate" is the system, if we can use such a word for a search engine that aims more for a kind of serendipity attracted by moods than for exact matches (how exact can be this kind of query, btw ?). I many be biased but I feel we need this kind of system at a large scale!
Like the mood option - very helpful. I personally also like it when a recommendation engine tells me why it thinks I will like the movie/restaurant/whatever. That way i can decide whether the recommendation makes sense for me instead of just trusting the "secret sauce". Thoughts on this?
It's really interesting the extent to which some people want to 'see the working' behind recommendations, while for others it's just clutter. I think some people want to build the mental model of what's going on, where others want it to feel like 'magic' (when it works, at least!)
I think my gut reaction against it is that people might reject a perfectly good recommendation because they didn't like the reasoning. For example, if we recommend something to Alex because Bob likes it, but Alex doesn't trust Bob's judgement, then Alex might reject the recommendation even if in practice she always enjoys the same things as Bob, she just doesn't realise it.
On the other hand, collecting feedback about whether the user agreed with the way the recommendation is made might be helpful in improving the way it works.
Agreed - Different users will have different reactions. Some will want to see the reasoning at least when they have just started using the service, just to see if the service works for them, while for others, it might be clutter. I like how Nara's iOS app does this (although it too doesn't do it perfectly).. they show you the recommended restaurants for you. When you click on any one result, they have a tab which shows you why they recommended it. That way, its there for the users who are struggling to make the leap of faith but not in the way of those who already trust the recommendation.
Also good point about collecting feedback about how users react to the way the recommendations was made. You could have different algorithms generating recommendations for different users depending on this e.g. collaborative filtering for some users, neural network for others, ensemble for the rest etc.
It would be nice if you could rate movies using shortcut keys (when hovering with the mouse pointer on a movie). That would be really helpful in rating lots of movies.
I assume there's some technical reason recommendations from Netflix, Amazon Prime, and Hulu aren't shown here? Is this localized for the UK or something?