You are here:  Home  >  Uncategorized  >  Current Article

Using Semantics to Solve the Weekend Movie Dilemma

By   /  October 9, 2009  /  No Comments

Jennifer Zaino
SemanticWeb.com Contributor

The long Columbus day weekend ahead may have you pondering how to spend your time. Movies are always a good idea — but how to figure out what to watch?

Jinni and its Movie Genome project want to help. Formally launched this week, the site aims at classifying movie and TV content beyond genre, and enables users to search for what they might want to watch using natural language. Jinni hopes to overcome what it says are the limitations of relying solely on recommendations determined by past personal content consumption and collaborative filtering of similar consumptions; searching content catalogues by filters such as genre; or through keyword searches.

“Jinni indexes meaning — that’s the semantic part of it,” says Yosi Glick, co-founder, president and COO. It understands, for example, that a movie like Little Miss Sunshine is about a dysfunctional family the same way as a human being reading a synopsis of it would know it was a story about a family that was a mess, he says. “Jinni understands that, and automatically gives it a semantic tag called dysfunctional family.”

The semantic technology actually surfaces the meaning of that particular content, Glick continues, extracting that based on the raw materials of synopsis and reviews. In addition to accounting for the “just the facts” element of a story, Glick says Jinni also uses sentiment-analysis techniques to determine within its database the magnitude of particular subjectives around a movie — for instance, whether it is a little humorous, medium-funny, or a laugh riot.

The movie genome, created by Jinni’s content team, defines more than 2,000 attribute vectors, and from that its algorithm then automatically assign some 30 to 50 semantic tags for each content item. In case you’re planning to see a movie a little farther out than this weekend, Glick says the database already can offer recommendations for movies due for release next January.

“We can do that because [our approach] is based on whether the story is the same,” he says. “As the parameters become more the same, the similarity is greater. We use semantic tags to calculate what is similar as opposed to what is consumed, and we measure that to be more accurate and we can start making the calculations even before any consumption takes place.”

Jinni is working its way backwards — new and more popular movies and TV offerings first, with the goal of eventually building a universal catalogue of all movies and TV shows. Its data comes from sources including affiliate partners and web crawlers.

The site is also planning to create a wiki to which users can contribute their own takes on content. That’s still in the works but Glick thinks it could be a mechanism for the community to voice its thoughts on whether it agrees or disagrees with Jinni’s interpretation of, say, a “feel-good” semantic tag.

“Maybe they think feel good is not the right way to say it,” he says, and if there’s a large enough sentiment on that point then Jinni might revisit its work and rename the tag. Or perhaps the community will identify tags that are missing and are worth adding to the database. In that way the community can function as a further quality assurance check on the work Jinni’s own content team does around defining and maintaining the catalogue of tags and algorithm output. Glick says the site measures 85 percent accuracy with the way its algorithms assign tags to content. He’s careful to point out, however, that it really will require a groundswell of community sentiment for Jinni to change catalogue or tagging practices. “That’s the difference compared to del.i.cious or Flickr where you can define your own tags as you wish,” he says. “Here it is a professionally defined tag and the generation is automatic. The users can agree or disagree and we can learn from that, but the tag structure is always professionally designed.”

Jinni will make its money as part of affiliate programs from partners such as Netflix, advertising, and at some point premium features that will require users who want them to pay a subscription fee. But perhaps the bigger opportunity is less as a destination site and more about using its technology to power content services for cable and web operators to enable their users to more seamlessly find the entertainment they prefer. It has one such deal so far, with Australia’s Quickflix DVD service.

You might also like...

Smart Data Webinar: Advances in Natural Language Processing II – NL Generation

Read More →