“Attacking the LEOs”: How to Protect Local Election Officials from Online Harassment

By Andrew Moseman

Image: Caltech

In three decades of studying American elections, Michael Alvarez, Caltech professor of political and computational social science, has seen online vitriol become an unfortunate but intractable part of the political process. During the 2020 election cycle, however, he says, hateful online speech reached a different, more personal level, one that has many election observers concerned on the eve of the 2022 midterm elections.

“What we saw in the last election cycle was, for the first time in my experience studying elections, a series of very targeted efforts to discredit election officials in the United States,” he says. “What I noticed was that many of these election officials who were targets of these attacks were people who, over the last 20 years, I've come to know very well personally and who we have worked with directly.”

As part of his research into topics such as the integrity of voting machines and vote counts, Alvarez has worked in close concert with the local election officials (LEOs) who oversee polling places and who allow Alvarez and his students to observe the happenings on Election Day. When those officials became the targets of online hate speech, he knew he could help. One of his newest political science projects, which launched with the moniker “Attacking the LEOs,” is a study funded by the National Science Foundation that uses artificial intelligence and machine learning to quickly identify attacks on election officials and to create instant actionable data for tech platforms or, potentially, law enforcement.

Alvarez had previously studied Twitter to identify hate speech injected into conversations around issues such as the #MeToo and Black Lives Matter movements. He knew his team could apply similar data mining techniques in this case, though the challenge at the outset was simply to identify the targets. No universal database of American state and local election officials exists, so his students set to work.

“They built some algorithms that essentially searched through Twitter conversations about elections and conversations among election officials. The algorithms then try to map out the network of election administrators in the United States, '' Alvarez says. “By then using that network information, we can build a directory of all the election officials in the United States and their Twitter usernames.”

Once the team built a data set of all tweets sent by or targeted at those accounts, they used a technique called topic modeling to sort the conversations by their subject matter and tone. Alvarez says his team’s tools have become very accurate at sorting positive or neutral tweets related to voting and election issues from tweets that are negative and threatening. With the potentially problematic tweets in tow, the team can then take a closer look to find any that rise to the level of harassment or personal attacks.

What the research found wouldn’t surprise observers of the current American political climate, Alvarez notes. “People who are disgruntled with the outcome of the election, people who are strong supporters of the losing party or strong supporters of a losing candidate, often take to social media to complain about the process, to complain about the outcome, to vent their anger,” he says. “Usually, that's just typical partisan conversation. Some of it is negative, but some of it starts to verge into trolling and harassment. Some of it starts to use profanity. Some of it starts to look a lot like a threat against an election official.”

Alvarez says social media companies have gotten much better and faster at identifying and removing harassment. In fact, Twitter often removes such tweets before he and his small group can find and catalog them. That’s a challenge for his research enterprise, he says—one that has led his team to develop tools to study Twitter in near real time. But, he adds, it is good news for democracy. It is both crucial to protect election officials from harassment and to counter misinformation about how voting and elections work.

“We have to get the facts out there about the integrity of the process,” he says. “We have to do everything we can to convince people that in the United States, elections are in fact being run freely and fairly and that they're free from fraud. That's a tall order for one faculty member at Caltech, but it is the sort of thing that here at Caltech we can do to help and that election scientists throughout the country are going to be doing in this election cycle and in the future to try to help turn this around.”