UCSD and CUNY researchers will analyze tweeted images to measure happiness.
A research consortium from UCSD and the City of University of New York Graduate Center was one of six teams to gain access to Twitter’s public and historical database for individual research purposes.
Last February, Twitter announced the other winning teams for its first #DataGrant program, including research groups from other countries such as Japan, the United Kingdom, Australia and the Netherlands. The only other U.S.-based team is from Harvard Medical School and the Boston Children’s Hospital.
The four-member UCSD and CUNY team of researchers is comprised of two UCSD alumni: data scientist Mehrdad Yazdani (Ph.D., 2012), who works for the Software Studies Initiative at Qualcomm Institute, and 2012 alumnus from the Interdisciplinary Computing and the Arts program Jay Chow. The other two research members are CUNY Graduate Center professor of computer science Lev Manovich and University of Pittsburgh history of art and architecture Ph.D. candidate Nadav Hochman, who is currently working under Manovich as a visiting scholar.
For their latest Twitter research project, the researchers will use open source tools, such as OpenCV and Python, and software they have developed themselves, called ImageMontage, to analyze selfies for features such as orientation and color measures.
The researchers have already concluded two similar previous projects known as Phototrails and Selfiecity, with datasets available online for public viewing.
Over 2.3 million Instagram selfies from 13 different cities were analyzed for the team’s Phototrails project in 2013. In addition, 3,200 Instagram selfies from over six different metropolitan areas were evaluated for Selfiecity in early 2014.
The researchers will be implementing the same method they used in their recent Selfiecity project to evaluate the Twitter selfies by looking for low to high level features.
“Low-level features, such as the brightness and saturation levels of images, provide us with an idea of the environment they were taken in,” Yazdani said. “While high-level features involve aspects like the person’s gender and the presence or absence of a smile.”
The visualization tools available at Calit2 will also be used to effectively evaluate the expansive database of selfies on larger displays.
Previously, research to measure happiness has largely focused on text shared on social media. The team hopes to extend the study to include images such as those shared on Twitter.
“There is so much value in data that we have not yet discovered how to learn all we can from,” Yazdani said. “This is why Twitter has reached out to the research community to see what people can come up with. I suspect that as we delve more and more into the project, new questions will pop up, and our data will provide statistically sound answers.”
Twitter views the creation of its #DataGrant program as a public service, as well as a method to promote the breadth of its database, which boasts an average of 5,700 tweets per second based on data provided by Twitter on August 2013.
The other winning teams will pursue research projects of their own, such as Japan’s converting Twitter data into a disaster informational analysis system and the United Kingdom analyzing how tweets play a role in the performance of sports teams.
Since Twitter’s announcement, Harvard Medical School and Boston Children’s Hospital have reached out to Yazdani’s group with the intention of starting a separate project involving the dataset from the Selfiecity project.
Manovich viewed the collaboration in a positive light for both parties to advance their research.
“The selfies and underlying data would be part of a global public health study,” Manovich said in an April 21 UCSD News Center online article. “We are delighted to start working with them because they are leaders in using social media for public health research.”
