Posts from the social media service Twitter are providing researchers at the University of Wisconsin-Madison a new way to study bullying among kids, the school said Wednesday.
UW-Madison researchers Amy Bellmore and Jerry Zhu, along with graduate students Junming Sui and Kwang-Sung Jun, have been able to teach a computer to identify tweets about bullying among Twitter's 250 million daily posts.
Zhu, a computer science professor, said many researchers are looking into data available through social media and many others are studying bullying.
"People haven't sort of put the two together," he said.
The researchers said that the computer was able to quickly learn to identify more than 15,000 bullying related tweets per day.
Zhu said the 15,000 tweets are just those that survived the filter applied to publicly available tweets.
"The number we have here is just the tip of the iceberg," he said.
Once the computer could identify bullying-related tweets, Bellmore and Zhu turned to using sentiment analysis to capture the emotions in the tweets. The computer hones in on strong emotional words used in the tweets.
The victims and witnesses in bullying incidents often expressed sadness or anger, Zhu said. The bullies themselves did not have many emotional posts, but when they did, Zhu said they would often be bragging.
The researchers found examples of both cyberbullying and discussions of real-world bullying, Zhu said. He added that cyberbullying was a smaller fraction of the results and there were more discussions of the aftermath of bullying in the physical world.
One goal of the researchers' work is to inform traditional bullying research. Typically, this research relies on self-report surveys in which victims and bullies report their experiences. This means researchers get a one-time glance at what is happening within a small population.
In order to inform traditional researchers, Zhu said, the Twitter data will need more calibration. Research is concerned with population level statistics, and so far there is not even a clear picture of the age distribution. With more refinement, the data could help traditional researchers.
Zhu also added that privacy is a major concern and the researchers make no effort to identify individual users.
Another goal is to be able to provide policymakers with better information about bullying. Zhu said the Twitter research could serve as the eyes and ears for policymakers by providing a better idea of the frequency with which bullying occurs.
Bellmore and Zhu also taught the computer to identify different roles played by kids. The computer was supposed to identify bullies, victims, accusers and defenders, but when the researchers looked into the bullying-related tweets they also found a new role - the reporter.
The other roles were identified in the 1990s, the reporter role is new, said Bellmore, an educational psychology professor.
"It's just like it sounds, a child who witnessed or found out about, but wasn't participating in, a bullying encounter," Bellmore said.
In some cases victims of bullying would change roles and end up bullying someone else, Zhu said.
The work could also be extended to other social networking sites such as Facebook or China's Weibo service.
However, adapting to a site such as Facebook presents other challenges. Zhu noted that Twitter posts are generally meant to be public, while many Facebook accounts are more private. Getting the data while maintaining privacy is a major hurdle, he said.
Bellmore and Zhu's work was presented at a computational linguistics conference earlier this summer and will also be discussed at a workshop in Beijing.
Most Popular Stories
- Slow Week Ahead of December FOMC Meeting
- Hispanics Seek to Grow School Board Members
- 'Knockout Game': Myth or Menace?
- U.S. Companies Eager for Iranian Business
- Questions Remain in Jenni Rivera's Death
- Banks Fret as Volcker Vote Approaches
- Entrepreneurs' Next Creation May Be New Laws
- GM Bailout Saved 1.2 Million U.S. Jobs, Report Says
- Bitcoin Used to Buy Tesla Car
- Paul Walker Fans Pay Respects