from this news site:


Computer spots a blockbuster from box office flop

LONDON (Reuters) - Hollywood producers fretting over this year's box office downturn should take heart.

A scientist in the United States says he has come up with a computer program that helps predict whether a film will be a hit or a miss at the box office long before it is even made.

"Our goal is to try to find oil in a way," said Professor Ramesh Sharda of the Oklahoma State University on Wednesday.

"We are trying to forecast the success of a movie based on things that are decided before a movie has been made," he told Reuters by telephone.

Sharda, an expert in information systems, has been working on the model for seven years and analyzed more than 800 films before publishing a paper which appears in "Expert Systems With Applications" journal early next year.

Sharda applied seven criteria to each movie: its rating by censors, competition from other films at the time of release, strength of the cast, genre, special effects, whether it is a sequel and the number of theatres it opens in.

Using a neural network to process the results, the films are placed in one of nine categories, ranging from "flop", meaning less than $1 million at the box office, to "blockbuster", meaning more than $200 million.

The results of the study showed that 37 percent of the time the network accurately predicted which category the film fell into, and 75 percent of the time was within one category of the correct answer.

Sharda said he was in discussions with a "major" Hollywood studio about further developing the system to make it more accurate. He did not name the studio.

Sharda may have picked the ideal moment to publish his findings.

As of mid-November, North American ticket receipts for the year so far stood at $7.6 billion, around 7 percent down on the same stage in 2004, although that was before the release of three big films: "Harry Potter & the Goblet of Fire", "The Chronicles of Narnia: The Lion, the Witch and the Wardrobe", and "King Kong".

His program only has a 37% accuracy rate?? I bet most of us on this forum can do better than that without using a complex model simulation.

Eventually, they'll get it down to a near-exact science but not anytime soon it seems.

It's not surprising for a computer program to do worse. There are so many human variables and so-called "buzz" that are hard to quantify, and one has to "feel" it him/herself.

The accuracy rate is not 37%; 37% of the time he will correctly pick the category; 75% of the time he will be within one category.

And xia is right, humans are much better at this stuff.