If you’re in the mood for a clever, contemporary movie about parents and children, the titles-only search on Netflix just won’t do. That’s where Jinni comes in. The “taste engine” matches users to movies via a variety of formats, ranging from specific keyword searches to a detailed browse mode.
After months of working with a psychologist to develop a movie-rating language made up of more than 2,000 tags in 20 categories, the Israel-based Jinni launched in November 2009. The site got some press last month, when Google named Jinni as a partner in Google TV.
I spoke recently with Jinni founder Yosi Glick. While he declined to talk about the Google venture, Glick explained the science behind Jinni’s complex system.
Jinni lets users search by key phrases to get movie recommendations, while Netflix only offers title search. Talk about that difference.
We believe we need to take a holistic approach for discovery, meaning sometimes we want to search, sometimes we want to browse and sometimes we want to get recommendations. It all depends on the cognitive effort we are willing to invest. If you are looking for something outside your comfort zone, you will use search. You come to the search box and type “feel good, smart, romantic movie in New York with unlikely couple” and in the results you’ll get As Good as It Gets. Jinni understands those natural metaphors, such as “feel good” and “unlikely couple.”
Sometimes, we want to browse. On Jinni, you click on Mood. It can be weepy, feel good, contemplative and so on. You can see what kind of movies fall under that. Later on, you can use the Squeeze Your Search refinement tool to further reduce the results. When you don’t feel like investing any time, you come to Jinni and you press Recommendation. Jinni will show titles that match you personality.
[By] giving a holistic experience, we are different than anybody else. Netflix and Amazon are statistical engines. They do not have the attributes such as “feel good” and “dysfunctional family” and they cannot support search. The only search they can support is when you search a title. Also, because they do not have the attributes, they cannot support browse. Statistical engines using collaborative filtering can use only one discovery method: recommendation.
Netflix devotes dozens of employees to personalization technology. How does Jinni, with only 20 employees, manage the complicated job of categorizing so many movies?
We’re using an automated process. All these tags are generated automatically. There is no human interaction. Most people would consider this magic, too good to be true. To say that Juno is [about] coming of age, this is a very detailed observation. We have very sophisticated technology that can identify all these details and generate the tags automatically. We don’t need hundreds of people. We just have the brain power to develop the algorithms. The algorithms analyze synopsis and reviews. From that, we generate the tags automatically. With that respect, we are very unique.
How does Jinni work with Netflix?
The whole idea is once you connect to your Netflix account from Jinni, practically speaking you don’t need to go again to Netflix.com. Within Jinni you can select content, put it into your queue, rate it, get more recommendations and so on. We provide an alternative for content selection for Netflix subscribers.
Image, top: Jinni
Image, bottom: Yosi Glick