Online discrimination linked to real-life hate crimes

It may be tempting to think that what happens online stays there. But by analyzing five years of Twitter posts from 100 U.S. cities, researchers at New York University have found that where race- and ethnicity-based discrimination was more common online, there were also more hate crimes in real life.

Building on this research may inform how social media companies control their content, and how police investigate hate crimes.

FBI data show that hate crimes have been on the rise across the country since 2014, and Philadelphia is no exception. Philadelphia police reports of hate crimes more than doubled from 2016 to 2018 — and the Philadelphia Commission on Human Relations suspects that many more incidents may go unreported.

Nationally, that discrimination has another home on social media. A study by the U.K. research group Demos found that in 2014, there were an estimated 10,000 racial slurs posted on Twitter in English — every day. Since the 2016 election, the number of discriminatory posts on Twitter has continued to increase, despite the website’s attempts to control this type of content. Twitter, along with other social media platforms, has been implicated as a venue for hate groups to safely grow, according to Data and Society, a nonprofit research institute.

It is unrealistic to separate these online and offline forms of discrimination, according to Rumi Chunara, assistant professor of computer science at NYU. “It’s becoming so pervasive. The internet is becoming more and more a part of our life — it’s part of the environment that we are immersed in," she said.

In a new study, Chunara’s team used a computer program to look at a random selection of 1 percent of publicly available posts from 2011 through 2016 — amounting to 532 million tweets. Using machine learning, the program identified and sorted two kinds of tweets with race- and ethnicity-based discriminatory content: those of people sharing their own experiences, and discriminatory speech targeted at others.

“Basically, the idea is to allow this [program] to learn a pattern without being explicitly told what that pattern is. Our [program] learns the properties of the tweets, and learns what words, or combinations of words, do or do not mean it’s a discriminatory tweet," she said.

The results showed that the greater the proportion of targeted tweets, the more hate crimes were occurring in the city. The association between online and offline hate held up even when taking into account the differences in population size, crime rates, and location of all 100 cities.

However, Chunara explained that the study only established a correlation between online and offline hate. How they might feed into each other is still unknown.

"In places where there are more hate crimes, does that enable social norms? Is there more discriminatory content and [then] people feel more comfortable elevating that to the level of crime? It’s a complex picture,” she said.

Svitlana Volkova, senior research scientist at the National Security Directorate at the Pacific Northwest National Laboratory, who was not involved in the study, said, “Social signals are not only predictive of offline phenomenon, but they can allow you to forecast the future. Imagine taking their findings [and asking] can I forecast the number of hate crimes tomorrow?”

Based on Chunara’s research, Volkova believes that scientists may be able to look at current social media and anticipate how likely discriminatory incidents might be in the near future. A predictive model like this could help police departments focus resources to help communities at risk, Volkava said.

For now, Chunara hopes that her research will encourage responsible parties to take action. “Research informs policy in some ways. There is obviously potential for larger governmental policies. And at the company level — they are paying attention,” she said.

Last week, Twitter updated its rules on certain kinds of hate speech. But the company continues to draw criticism based on its handling of racially charged content from high-profile users such as President Donald Trump.

However, the connection between online and offline behavior suggests that the current hate-filled state of social media need not be the future. “We change our social environment just as our social environment can change us," Chunara said.