Tweets aren’t big fans of #oscar nominated films, @princeton team finds

A research paper released last month by Felix Wong GS and postdoctoral research associate Soumya Sen of the electrical engineering department demonstrated that Twitter users are both more positive in their reviews than other movie rating sites and that tweets tend to be less positive in their reviews of Oscar-nominated films than non-nominated ones. The paper also questioned the ability of Twitter to predict success in theaters.

On Saturday, Wong and Sen learned that their paper was accepted to a workshop at SIGCOMM, the flagship annual conference of the Association for Computing Machinery.

Although the two have been working on large scale data analysis projects like this one for over a year, their study of Twitter first took form in January, when electrical engineering professor Mung Chiang suggested Twitter as “the medium to monitor,” according to Wong.

“[Chiang] has a lot of vision,” Wong said. “He gave very specific suggestions on what to do, the scope of the project, how to write the paper. [Sen] suggested tracking movies, and I discovered which tools were applicable to the analysis,” he explained.

“We realized that if we don’t collect now, we won’t get the Oscar-related data, so we then hurriedly started,” Sen said.

From early February to mid-March, Sen and Wong collected data from 12 million tweets by tracking mentions of 34 movies. They also collected Rotten Tomatoes and IMDB scores for each movie. They used a classification system called the support vector machine that labeled the tweets in certain ways. Each tweet was then ranked in three categories the researchers had developed: relevance, sentiment and time frame. The tweets were marked as relevant or irrelevant; negative or positive; and before, after or during the movie.

Sen and Wong manually labeled 11,000 tweets so that the SVM could learn by example. In addition, they created a new set of metrics to measure the positivity, bias and “inferrability” of tweets, as well as a measurement known as the hype-approval factor for a movie, which is the ratio of the number of tweets about the movie before its release to number afterward.

“We developed our certain kind of metrics using some information theory-oriented approaches,” Sen explained. “We wanted to figure out, how can we infer the true predictive ability of movie quality? If you know how well the Twitter population seemed to like it, how can you predict the IMDB score?”

Wong and Sen found that the number of positive reviews on Twitter far exceeded the amount of negative reviews, a discovery that could impact general marketing strategies, they said.

“Part of this could be from the culture of Twitter — to your followers, you want express yourself as a positive person rather than complaining all the time,” Sen said. “Those may be some psychological aspects we can look at.”

While Twitter reviews of recent releases were generally positive, those for Oscar-nominated films were slightly more negative. However, these Twitter ratings, along with hype-approval factor, usually could not predict the rating from the general online population.

Lastly, Sen and Wong found that box-office predictions were just as difficult to make. While no movie with high hype and rating was financially unsuccessful, some with low hype and a low rating could still perform well at the box office, while those with mixed ratings and hype could perform either way.

Get the best of the ‘Prince’ delivered straight to your inbox. Subscribe now »

The researchers said they hope their work demonstrates the diverse research opportunities the electrical engineering department offers.

“It’s usually not an [electrical engineering] thing, it’s usually a [computer science] thing,” Wong said of the interdisciplinary aspect of this research. They said they hope this will get rid of common misconceptions regarding the department.

“Students should know it’s not just about traditional things like building chips and electric circuits. There’s a lot of fun in analyzing data sets and carrying out real world trials,” Sen said. “It’s not just core lab, which students typically associate with EE. It’s a much broader field.”

Network research also opens up many multi-disciplinary research initiatives. With ties to sociology, economics, psychology and computer science, its use in predictive economics and elections has generated media buzz, with the results of the paper becoming the subject of articles in Politico and the Wall Street Journal.

Undergraduate students will have a chance to learn more about the popular topic. Chiang, a third author of the paper, is again teaching ELE 381: Networks: Friends, Money and Bytes, a course centered on the social media platform that highlights social, economic and technological networks.

“Undergrads these days grew up in the age of Twitter and the iPhone,” Chiang said in an email. “They can readily appreciate and even carry out research like Twitter movie analysis. A lot of interesting implications can arise out of a large volume of empirical data plus quantitative reasoning.”

“Especially now with all this social media, we have a huge amount of data sets right in front us, easily available,” Sen added. “You can tap into what people say in real time. You can track lots of interesting research questions.”

Though they said they were pleased with the response their research has gotten, Wong explained that the researchers are not focused on the attention from the outside.

“It was not something we sought,” Sen added. “We are more focused on tying up the loose ends and carrying it forward. There is a lot of work to be done.”