By Kasra Zarei
The U.S. film industry is worth billions of dollars, with hundreds of Hollywood movies made every year. The secrets to making profitable movies have evaded producers and investors for decades. Now, two University of Iowa researchers have developed an algorithm to predict the success of box office movies in terms of profitability.
The research of Kang Zhao, a UI assistant professor of business, and Michael Lash, a UI graduate student in computer science and the lead author on the project, could present a solution to this problem that film studios, financial analysts, and other academic teams have unsuccessfully tried to answer for years.
Zhao’s research involves data analytics for social and business networks, with his most recent work in collaboration with Lash focusing on predictive modeling of box-office movie success.
“We looked at factors that drive the success of movies from a targeted perspective, being profits,” Lash said.
By defining novel features of movies and appropriately leveraging them, Zhao and Lash have been able to make pre-production predictions of profitability with better than 90 percent accuracy on historical data.
“One of the approaches we took was an interpersonal, connected look at people who are involved in making movies,” Lash said. “That is, historically, what level of profit have people seen in their past collaborations.”
This research question has been a large undertaking for Zhao and Lash, who have worked on the project together for almost three years.
“The first steps of the project were to acquire historical data,” Lash said. Using extracted data from various sites including Imdb and Box Office Mojo, Zhao and Lash iteratively defined the features or attributes of movie success with a profit-centered focus.
Zhao and Lash ran a set of predictive algorithms that use features of movies that influence profitability like big name actors, movie genres, plot synopses, and release locations to identify patterns in past data.
“We wanted to find the best predictive model from a global set of classifiers that exists,” Lash said.
Among all the various approaches tested, Zhao and Lash found that their top performing classifier consisted of a “Random Forrest Classifier,” which incorporates a sense of randomness, feature selection, and voting in the profitability decision-making process.
“[Predicting movie success] is a historical problem,” Lash said. “People have been looking at this problem for years, getting data from the Internet and running computations on machines.”
However, what differentiates Zhao and Lash’s research is the way they define movie success.
“Oftentimes, people define success in terms of box-office gross, but this is in no way indicative of boxoffice profitability,” Zhao said.
But their recently conducted study has found a loose correlation of gross revenue with profit in the film industry.
“While big-name stars may drive the box-office revenue generated, the price tag associated with these stars may not equal profitable returns,” Lash said.
Compared with previous approaches that assess movie success using post-production information, Zhao and Lash use a more realistic approach.
“We define our features, or variables, carefully such that we are not using any information that is available right before or after release,” Lash said. “We only use features that are available during the investment stage of a movie so that we can actually help investors make their decision prior to production.”
Lash hopes to take a more targeted look at how they can help the different roles of people involved in making a film.
“For instance, if you are a writer, we want to know what we could suggest to you to make your script better,” Lash said.
Nick Street, a UI professor of management sciences, said this kind of work will be important to solving a variety of issues in different areas.
“It is a great example of the highly interdisciplinary work going on in the Management Sciences Department,” he said. “[They are] applying methods from computer and information science to problems in business, social sciences, biology, and health care.”
