San Diego, CA, March 07, 2014 -- Networks are not what they were ten or even five years ago. As networks change, so do the ways in which they can be studied. Engineers at UC San Diego and around the world are beginning to applying rigorous scientific methods to a range of new and emerging networks – including the networks of people who use the Internet.
The headline-grabbing, UC San Diego-led study “Detecting Emotional Contagion in Massive Social Networks” from March 2014 serves as an example. The two electrical engineers from UC San Diego who participated in the study—first-author and Ph.D. candidate Lorenzo Coviello and professor Massimo Franceschetti work at the cutting edge of a new engineering subfield called “network science.”
|Ph.D. student Lorenzo Coviello (left) and professor Massimo Franceschetti (right) are from the Department of Electrical and Computer Engineering at the UC San Diego Jacobs School of Engineering.|
Coviello and Franceschetti recently sat down and offered a bit of context for this project, which is well summarized in the UC San Diego press release: “Facebook Feelings Are Contagious, Study Shows.”
The Networks of People Atop the Internet
Engineers have already worked out a lot of the issues related to how the Internet functions. Researchers have also uncovered many new questions and challenges that the Internet itself raises for network science.
But there are additional layers of information in online networks to study, including the networks of people who use the Internet. How do the people on top of the Internet interact? Scholars in engineering, mathematics and physics are all looking into these kinds of questions, using the tools of mathematical modeling, statistical analysis, data mining, network theory, network effects and more.
In the recent social contagion in massive social networks study, for example, researchers analyzed over a billion anonymized status updates among more than 100 million users of Facebook in the United States. Positive posts beget positive posts, the study found, and negative posts beget negative ones, with the positive posts being more influential, or more contagious.
To identify possible causal relationships in the massive data set, the researchers needed the find a mathematical model capable of representing the data. As a starting point, they looked to econometrics, a field that has used “instrumental variable regression” models to uncover causal relationships.
“Our goal was to provide a model to capture causative effects using observational data. This was the main problem we were facing at the beginning of the project,” said Coviello.
“Lorenzo Coviello developed a model based on an external variable that influences the emotions of people—rain,” said Franceschetti.
In this way, rain served as a type of scientific experiment.
“We cannot control weather, but if it’s a rainy day, it’s like we are running an experiment,” Coviello said.
Using this model, theresearchers analyzed the billion updates among more than 100 million users of Facebook in the 100 most populous U.S. cities between January 2009 and March 2012.
The researchers report that rainy weather reliably changes the tenor of posts, increasing the number of negative posts by 1.16 percent and depressing the number of positive by 1.19 percent.
The researchers used the rain-induced change on the emotional content posted by a user to estimate the spill-over effect on her friends who live in different cities and are not experiencing the same weather.
“Our study suggests that people are not just choosing other people like themselves to associate with but actually causing their friends’ emotional expressions to change,” said lead author James Fowler, professor of political science in the Division of Social Sciences and of medical genetics in the School of Medicine at UC San Diego. “We have enough power in this data set to show that emotional expressions spread online and also that positive expressions spread more than negative.” (Quotes are from the original press release.)
Causation vs. Correlation
While Coviello and Franceschetti played a key role in developing the models that teased causation out of the Facebook data set, these electrical engineers highlight the fact that they did not come up with the final solution to the correlation-vs-causation question. That question will be around for a long time. Instead, the model used in the recent study is the best one that the team could generate for this specific data set, given current technology.
“Maybe in two or three years we will have a better solution. But using the best technology today, this is our way to solve the problem,” said Franceschetti.
And in two or three years, likely even more people will be studying the nature and behavior of massive social networks. In addition to the possible implications for public policy and public well being, these are the networking questions that many of the undergraduates in Franeschetti’s classes want to study.
And as this discipline of network science further develops, researchers will have the tools to study more kinds of new networks is rigorous ways, which in turn, will open up even more possibilities for research collaborations between engineers and social scientists.
The research described in “Detecting Emotional Contagion in Massive Social Networks” was partially supported by Army Research Office Grant W911NF-11-1-0363, and grants from the National Institute for General Medical Sciences (P-41 GM103504-03) and the Pioneer Portfolio of the Robert Wood Johnson Foundation.