-Semantic Technology Delivers Informed Recommendations
-New Web Site, Mobile Device App Coming
Yosi Glick, founder of the Israel-based video discovery and
recommendation engine Jinni, told The Online Reporter that in
terms of video discovery engines, the race is still wide open.
He said all of the current search and recommendation engines
rely on “traditional metadata, which [has been] available for
the industry for the past 60 years.”
“That alone is not good enough to build true discovery,” he
His recommendation engine and its Movie Genome Project works
with two cable operators in the US, the Best Buy smart TV app,
and Xbox live, and Belgacom, Swisscom and Prisa in Europe. It
also has established a wide fan base among movie viewers on
“Fundamentally, we can see the market has two kinds of
engines,” he said. “Statistical engines that are using the
famous collaborating filter and engines that use the technique
of understanding what the product is all about,” which is
called semantic technology.
Glick said Jinni identifies three relationships: the
relationship between titles, the relationship between user and
content, and the relationship between users.
“When we started Jinni, I started it with a filmmaker and a
psychologist,” Glick said. He said Jinni is tried to develop
an engine that understands content “that is as close as
possible to the way human beings think and act.”
“If we want to think about the concept of collaborative
filtering, I do not know anything about the attributes of the
product,” Glick said. “ ‘People who watch this also watch
that’ – that is the [foundation] of collaborating filtering.”
“When we [humans] think about product, we don’t think
statistically,” he said.
The Power of Semantics
The Jinni engine offers users a number of topics to include in
the search, and encourages users to list attributes from
different topics in the search. The Movie Genome project has
2,200 parameters. “We have a very wide set of attributes that
we define,” Glick said. The algorithms process synopses and
reviews of the films.
-Mood, such as bleak, clever, exciting, tense and stylized.
-Plot, including family life, good versus evil and criminal
-Genres, such as crime, thriller, sci-fi
-Time/period, from ancient times and the Middle Ages to the
-Place, which includes Paris, high school, small town,
-Audience, such as teen, girls’ night and date night
-Praise: Cannes, Oscars, Cult, Blockbuster, among others
“Jinni is looking for meaning in synopsis,” he said. “We
create a common language across all titles, irrespective of
the language used in the synopsis,” which are provided by
Tribune Media Services. The Jinni technology processes those
titles, and in essence translates the meaning into the
engine’s own framework – ie whether a film belongs in “bleak,”
“witty” or “atmospheric” mood categories.
For example, Glick said the technology can determine from a
synopsis that a movie is about the rise and fall of a
character without the words “rise and fall” being used.
“We want to define the attributes of the title,” Glick said.
“Jinni is reverse engineering of the intention of the
Getting Recommendations Right
Glick said the Netflix method of cataloging and recommending
movies based on one attribute is not the best way to recommend
content. Once a viewer has watched a World War II movie, for
example, Netflix will continue to suggest WWII-themed films.
“It creates for you a long list of titles that are good for
you, which is ok,” Glick said. “From the consumer point of
view though, ‘am I in the mood?’ ” he said.
Jinni’s recommendations are not just based on movie
attributes; they are also based on viewer attributes.
Jinni encourages users to develop a “movie personality” by
rating a bunch of movies. Using this indicator, Jinni will
keep track of what a user watches and how the user rates it.
By giving users with a much wider vocabulary than simple, one
attribute ratings, the engine is ultimately able to deliver
much more informed recommendations.
“Jinni is able to find who [a user is], like an x-ray,” Glick
said. “That really delivers the wow-factor.”
Jinni also allows users that utilize social media connections
to leverage video discovery. The social component of the
service is highly personalized – whereas most recommendation
engines that incorporate social networks tend to focus on what
is popular or trending among friends, Jinni allows users to
determine which friends are most relevant to the user. That
allows the engine to make, once again, more informed
recommendations. Finally, Jinni also pairs Jinni users with
one another based on movie personalities, allowing users to
connect and even recommend titles to one another.
Glick said Jinni will be launching an updated Web site with
more advanced social features at the end of the year, and will
release a mobile device app in early 2013.
“You said that humans are better at making recommendations
than machines,” Glick told us, “When in fact it is the exact