Compute Me The Money


Nov | Dec 2010 contents
Gazette Home



Making History campaign reaches $2.85 billion

Northwestern’s J. Larry Jameson named new medical dean and EVP

Thouron program marks 50 years

Synthetic life “not a novel thing” says Bioethics Commission witness

Convocation: Passion and purpose drew Penn to Class of 2014

Silicon Valley moves east for Supernova conference

Ei-ichi Negishi Gr’63 shares Nobel Prize in Chemistry

Screening for cancer with vinegar and a cellphone camera

Penn Guardian offers GPS location in emergencies

A grandson’s memories of Eisenhower in retirement

Wharton model for movies gives “formulaic” a whole new meaning

Football-induced brain injuries possibly linked to student’s suicide


Bagnoli becomes Penn’s winningest football coach


Share |



Casablanca. High Noon. Chinatown. Each film has won Oscars, critical plaudits, and decades of audience praise. Even without seeing these movies, people know their most famous lines. At the first script reading of Casablanca, though, the producers probably thought it was just another movie. Most likely, no one dreamed that people would still be saying “Round up the usual suspects” 60 years after the premiere.

Or was it predictable? Jehoshua Eliashberg, the Sebastian S. Kresge Professor of Marketing, and John Zhang, the Murrel J. Ades Professor, working along with Sam K. Hui of New York University, have developed a method to forecast a movie’s success based on computer-aided script analysis. Using a combination of “domain knowledge, natural language-processing, and statistical learning methods,” they are able to predict box office revenue, usually to within $15 million. (The average Hollywood movie costs about $90 million to make and market.) Eliashberg’s three favorite movies—the ones listed above—would make the cut, he says.

As baseline data for their predictive model, the team analyzed the frequency of certain words, themes, and characters in written dialogue in more than 200 scripts. Then three movie experts read each script and answered 22 questions about them, such as whether it featured a “surprise ending” or a “multi-dimensional conflict.” In a recent paper, the researchers noted that by using their instrument to select the most promising 30 scripts out of 81 movies actually released between 2001 and 2004, studios could have achieved a return-on-investment of 5.1 percent—compared to a loss of between 18 and 24 percent that would have resulted from green-lighting 30 of those scripts chosen at random. So the model shows some monetary promise. Yet “bringing the science” to movies, as Eliashberg puts it, has had mixed results with people.

According to the professor, who calls himself a “movie freak,” many in Hollywood think his research team has “lost it”—while some groups want more. Their model has sparked favorable interest among “private equity firms [and] independent movie producers,” Eliashberg says.

There are also skeptics. Kathleen Van Cleve, a senior lecturer in Penn’s cinema studies department, told the Daily Pennsylvanian: “The idea of basing creative decisions on any kind of mathematic formula (no matter how extensive) makes me shudder.”

But though one might think this sort of predictive formula might alienate Hollywood’s creative types, Eliashberg says screenwriters have shown some of the greatest interest. After all, writers have to master a tricky balancing act. They want fresh movies that are different enough to stand out, but they also want someone to actually produce their script. But even so, judging creative work with a species of model once used in “engineering to predict the quality of glass coating,” as Eliashberg says, seems a little strange. One wonders if reliance on such a tool might lead to overly formulaic writing.

Then again, perhaps the concept is not so foreign. Deborah Burnham G’76 Gr’79, associate undergraduate chair of Penn’s English department, points out that sometimes a strict formula or structure can lead to “wonderfully luminous” works. Sonnets, after all, have adhered to very precise rules of meter and rhyming for several centuries. Burnham, a poet herself, notes that those formulas served Shakespeare and John Donne quite well. Yet “the question that this raises,” she adds, “is, Do you cut off imaginative possibilities if you think that a [certain type] will sell?”

Then again, studio executives already do that on a daily basis. Perhaps Eliashberg’s predictive model would at least rescue them from green-lighting the next Ishtar.

Laura Francis C’13


©2010 The Pennsylvania Gazette
Last modified 10/25/10