Can an Algorithm Make Sense of Readers’ Reactions?
by Chris Carroll | illustration by Steffanie Espat '15
We’ve all read a fascinating article online, only to stumble into a dark forest of off-topic, ALL-CAPS name-calling in the comment section.
But UMD researchers are designing a system that uses computer learning to help editors quickly mine through hundreds or thousands of comments, pushing the most relevant and well-written to the top of the heap.
Nick Diakopoulos, an assistant professor and head of the Philip Merrill College of Journalism’s Computational Journalism Lab, studies comment moderation at The New York Times, which employs 13 full-time comment editors. The system he’s developing, CommentIQ, aims to give lone editors at smaller publications the same level of oversight.
An algorithm does the initial read, quickly scoring comments on editor-set parameters like readability and the amount of personal experience reflected. Crucially, CommentIQ doesn’t make the final cut, but delivers a winnowed-down selection of relevant comments so a real person can highlight the best of the best.
“It takes advantage of what computers are good at, and what people are good at,” says Diakopoulos, who’s also a member of the UMD Human-Computer Interaction Lab. “We’re interested in the sweet spot between automation and human judgment.”
His collaborator, Niklas Elmqvist, an associate professor in the College of Information Studies with a joint appointment in the university’s Institute for Advanced Computer Studies (UMIACS), says publications have much to gain from civil, intelligent comment sections.
“In an ideal world, people have a fruitful exchange of ideas, although in reality, that’s usually not what happens,” he says. “But if the comments become valued additions to the article, you build a community of commenters and engage readers.”
