Shrikanth “Shri” Narayanan isn’t much of a movie buff.
The professor of electrical engineering and computer science at the USC Viterbi School of Engineering is more of a musician. He plays classical music on a veena, a plucked string instrument that originated in his native India.
But in an intriguing research project he conducted with two of his lab members, Naveen Kumar and Tanaya Guha, and an intern, Adarsh Tadimari, last summer for USC Viterbi’s Signal Analysis and Interpretation Laboratory (SAIL), Narayanan dove into one of Hollywood’s most recognizable marketing tools: the movie trailer.
The two-minute-or-so teaser is a key tool in movie studios’ marketing arsenals as they try to put as many bodies as possible into the seats of cinemas during the most important period for a new movie, the opening weekend. In fact, movie studios typically spend about 4 percent of their marketing budgets on trailers.
Grosses on opening weekend can spell doom or success for a new flick. Predicting initial ticket sales is a longstanding practice that traditionally has relied only on “metadata” such as the lead actors in the movie (Tom Cruise is a safe bet), run time (a three-hour movie poses challenges), genre, number of screens, MPAA rating (an NC-17 is a big hurdle), pre-release buzz and Google search trends.
But what about movie trailers? Can the content of trailers also help predict box office performance?
Using computers to perform the statistical technique of linear regression analysis — the study of linear, additive relationships between variables — Narayanan and his team analyzed trailers for 474 American movies released during from 2010 through 2014, from thrillers, slasher flicks and action movies, to comedies and romantic movies.
The team extracted audiovisual features from trailers including the number of close-up shots of actors and the predominant hue (blue indicates darker movies; orange, warmer ones), and then put the information into a database to capture emotion-related information in
“It’s a Holy Grail question: How can content predict the potential impact on people, whether we’re talking commercial or societal impact?” said Narayanan. “This study is one baby step in connecting content to impact. We wanted to take a crack at determining whether content gives us any insights into box office success. And what is it about that content that connects with people? That’s a bigger question we are trying to ask.”
The movie trailer study is part of a larger research effort at USC to analyze media content, using data science techniques familiar to engineers and computer scientists, such as for insights into gender and race bias — unconscious or not — in entertainment.
“We found that the audiovisual features alone in movie trailers significantly correlated with some of the metadata features in predicting opening weekend grosses,” said Naveen Kumar, a Ph.D. student who co-authored the study, “Opening Big in Box Office? Trailer Content Can Help,” which was presented in March at a conference in Shanghai.
For example, the study found that audiovisual features in trailers are better at predicting opening-weekend box office than movies released in the genre of science fiction. And a trailer’s audiovisual features are nearly as good of a predictor of success as a sequel to a popular film.
Metadata analysis alone has a 60 percent explained variance in accurately predicting weekend grosses. Trailer content has an 11 percent explained variance. When combined with metadata, the content of a trailer can improve by 6.2 percent the model’s strength in predicting opening weekend box office.
“This study is another tool for movie makers to think about when it comes to the role creative content plays in the trailer,” Narayanan said.
For the record, the professor, who rarely has time to catch a flick, has a favorite:
“My Cousin Vinny,” the 1992 comedy starring Joe Pesci and Marisa Tomei, who won an Academy Award for Best Supporting Actress. The movie’s opening weekend gross was $7.4 million, good for No. 2, according to boxofficemojo.com.
“What we attempted to do with this study was ask what role trailers play in box office success, and if they do, how we can make use of trailers in a particular way,” Narayanan said. “Whether we can make quantitative something that is more fairly qualitative still is an open question.”
Stay tuned for the sequel, coming soon to an engineering lab near you.