Researchers have created a model to predict which civil online conversations might take a turn and derail.
After analyzing hundreds of exchanges between Wikipedia editors, the researchers developed a computer program that scans for warning signs in participants’ language at the start of a conversation—such as repeated, direct questioning or use of the word “you”—to predict which initially civil conversations would go awry. (Edtor's note: For info about the online quiz, see the end of this article.)("Guess which conversation will go awry" online quiz: http://awry.infosci.cornell.edu/)
Early exchanges that included greetings, expressions of gratitude, hedges such as “it seems,” and the words “I” and “we” were more likely to remain civil, the study found.
“There are millions of such discussions taking place every day, and you can’t possibly monitor all of them live. A system based on this finding might help human moderators better direct their attention,” says Cristian Danescu-Niculescu-Mizil, assistant professor of information science at Cornell University and coauthor of the paper.
“We, as humans, have an intuition of whether a conversation is about to go awry, but it’s often just a suspicion. We can’t do it 100 percent of the time. We wonder if we can build systems to replicate or even go beyond this intuition,” Danescu-Niculescu-Mizil says.
The computer model, which also considered Google’s Perspective, a machine-learning tool for evaluating “toxicity,” was correct around 65 percent of the time. Humans guessed correctly 72 percent of the time.
People can test their own ability to guess which conversations will derail at an online quiz.
The study analyzed 1,270 conversations that began civilly but degenerated into personal attacks, culled from 50 million conversations across 16 million Wikipedia “talk” pages, where editors discuss articles or other issues. They examined exchanges in pairs, comparing each conversation that ended badly with one that succeeded on the same topic, so the results weren’t skewed by sensitive subject matter such as politics.
The researchers hope this model can be used to rescue at-risk conversations and improve online dialogue, rather than for banning specific users or censoring certain topics. Some online posters, such as nonnative English speakers, may not realize they could be perceived as aggressive, and nudges from such a system could help them self-adjust.
“If I have tools that find personal attacks, it’s already too late, because the attack has already happened and people have already seen it,” says coauthor Jonathan P. Chang, a PhD student at Cornell. “But if you understand this conversation is going in a bad direction and take action then, that might make the place a little more welcoming.”
The paper, co-written with additional collaborators at Jigsaw and the Wikimedia Foundation, will be part of the Association for Computational Linguistics’ annual meeting (July 2018) in Melbourne, Australia.
Guess which conversation will go awry!
Online Quiz Instructions:
In this task, you will be shown 15 pairs of conversations. For each conversation, you will only get to see the first two comments in the conversation. Your job is to guess, based on these conversation starters, which conversation is more likely to eventually lead to a personal attack from one of the two initial users.
After answering each question you will get instant feedback on whether your answer was correct (indicated with green) or incorrect (indicated with red).
In making your guess, you should use the following definition of a personal attack as reference:
A personal attack is a comment that is rude, insulting, or disrespectful towards a person/group or towards that person/group's actions and/or work.
Keep in mind that you are not looking for personal attacks in the comments that are shown. Rather, you should be using your intuition of social dynamics to decide which exchange is more likely to lead one of the participants to eventually post a personal attack (which you are not shown).
At times, it might seem like neither quote is likely to lead to an attack, or that both seem equally likely. However, please keep in mind that the source conversations have already been annotated by humans, and one indeed leads to a personal attack. Do your best to 'recover' those existing labels!
This is not an easy task, and it might take a couple of minutes to answer each question. As this is a difficult task, the first three questions are "warm-up" questions that will not affect your score; they are there to help you "calibrate" your sense of what factors are likely to signal future attacks. But remember, your task is to recover as many labels as you can.
Due to the nature of the task, some of these comments might contain offensive content. We are sorry about that.
Source: Cornell University