Finding a movie to watch on a rainy Friday night can be like finding the proverbial needle in a haystack.Jinni is a content discovery system, or as the makers prefer to call it a “taste engine”, for movies and TV shows which addresses this problem. The service has just gone into public beta.
Most of us choose movies based on rather amorphous criteria like mood or an association with another movie we like. Categorisations like genre are too wide; titles are too specific. Jinni approaches discovery in an intuitive way. You can search for movies and TV shows based on mood terms like “witty”, “stylized” or “disturbing” or plot elements like “unlikely couple” or “ambition”. The results are presented visually (we are seaching video after all) with more popular results getting bigger images. You can also tune results to request lesser known titles or faster paced content.
Links are provided to sites like Amazon, Netflix and iTunes (and for European users LOVEFilm) where you can buy or rent the content. The recommendation system learns rapidly based on ratings you give for the movies you find in the first searches. Within a few minutes it was coming up with recommendations for unfamiliar content as well as movies I already knew and liked. Jinni expects to offer high quality recommendations after 10-20 ratings. The database currently contains 25,000 titles, 20% of which are non-English language.
What’s New Pussycat?
The public beta adds several new features. The most important is the movie personality sketch which claims to identify each user’s unique entertainment personality. These are the parameters which are most common in your choice of movies and are presented in a word cloud with examples of typical movies with these characteristics. You can then jump to other titles which share the same characteristics.
If you click on another user, you will see your shared tastes like “You both especially like clever stories about betrayal and gangsters” as well as movies you both like. Other community features like social recommendations and the pulse (what’s happening on Jinni right now) have also been introduced.
The Matrix (aka the Science Bit)
The most common recommendation method used by web sites like Amazon is collaborative filtering. This groups users who are considered to be similar to each other and assumes that if other users in the group liked something then you should also like it.
The core of Jinni is a “movie genome” similar to Pandora’s music genome, which categorises video content using 2200 different parameters or “genes” covering mood, style, setting, atmosphere, etc. Natural language processing and sentiment analysis to used to derive the genes from reviews and other information available online about the movie. The combination of algorithms used to extract the genes, determine their relevance and use them in recommendations is Jinni’s golden egg and a closely guarded secret. This is not an easy problem given the subtleties of language and meaning (“Is dark humour the same as irreverent humour?” “What is the distinction between assassin, serial killer and slasher?”) not to mention the unorganised nature of the web data which is used. It’s refreshing to see a startup which is actually tackling a difficult technical problem. Crowdsourcing is not the answer to everything.
Jinni has several advantages over other recommendation engines. As well as offering good recommendations quickly, it can explain why a particular movie was recommended based on the most relevant genes and even point out elements of your movie taste of which you may not have been aware. In contrast, collaborative filtering can only tell you that someone who has similar taste to you liked the movie.
A minimum of 4 reviews is required to analyze a title and even long-tail titles have typically been reviewed many more times. This method means that a title can be indexed even if it has never been watched by a Jinni user.
Jinni’s most obvious rival is Netflix, which just awarded a 1 million dollar prize for improvements to its recommendation algorithm. The founders of Jinni see the two services as complementary with Netflix’s main focus being content delivery while Jinni does discovery. In fact, Jinni has already integrated with the Netflix API due to user demand. I don’t entirely believe the argument that Netflix is not a competitor. It seems unlikely that Netflix would place such emphasis on its recommendation engine if it was not seen as a fundamental part of the business.
There are many other recommendation engines (e.g. Tastekid, Clerkdogs, Criticker) but none of them seems to cover all the functionality available in Jinni or match its pleasing design. The closest is spirit to Jinni is probably Nanocrowd which uses its own 3 word genome called the nanogenre. An example is happiness-struggles-pretentious which apparently applies to “The Hours”.
Dirty Sexy Money
The Jinni team believes that just as most content on the web is free, while filters like Google rake in revenue, video content will become the commodity and personalised discovery the value-add. The business model therefore involves licensing the Jinni API to TV operators and Internet content providers as well as advertising based on user taste modelling. There are also plans to introduce a premium package with personalization features by mid 2010. In the longer term, Jinni wants to apply the genome approach to other products like books and video games.
We need to talk about your flair