|
No it's not a crystal ball, but it comes close to one in predicting whether a movie will be a hit or a miss. It's called "Movie Forecast Guru" and is the brainchild of two Oklahoma State University professors, Ramesh Sharda and Dursun Delen. This neural network based model takes movie attributes, like star power, technical effects, the genre of the movie, the month of release, number of screens and puts them together into a nonlinear model, which tries to forecast the success of the movie in one of nine categories. The lowest category is super flop and the top super success. It can predict exactly how much revenue the movie will generate 37 percent of the time and provides accurate predictions 75 percent of the time! Ramesh Sharda chats with Little India.
Q: What is the Movie Forecast Guru?
A: It's a computer program that learns from past behavior and it uses this technology known as neural networks, which is used for many other forecasting problems. It just happens we are probably one of the first ones to use neural networks for forecasting box office performance. Otherwise neural networks have been used for forecasting the failure of corporations. My earlier research was on the likelihood of firms going bankrupt. Based on its historical performance, we could predict quite successfully whether a firm would be around next year. That idea is used in all our credit mortgage loan approvals.
Q: Was this totally a fun project for you?
A: Yes, it was a fun project. I've used neural networks for many serious problems like business decision problems.
Q: Do you have to actually sit through all the movies?
A: No, our data is gathered from movie databases, publicly available data sources which we feed to the model. We use it to train the model and then a different set of movies, which it has not seen before, is fed into it to test whether it's learnt anything.
Q: Are you thinking of patenting this?
A: We are trying to figure out what is the best way of commercializing it . Our primary approach will be to work with studios to refine the model because the studios have access to many more parameters and we want to see if we can improve the performance of the model even further by incorporating more details about the movie. So it's just a research step at this time.
Q: Are you yourself a movie fan?
A: Yes, of course. We see a Hindi movie every week and an English movie every couple of weeks.
Q: So Bollywood will be very grateful to you if this takes off?
A: Interestingly, while major Hollywood studios have contacted us, so far I haven't heard from anyone in Bollywood, although the story was carried in major Indian newspapers. There would have to be clearly additional variables that would have to be incorporated for Bollywood films - such as the number of songs and the number of costume changes! | 
