In a society bombarded with information, Western Washington University Computer Science graduate student Ellyn Ayton of Tacoma is working to detect some of the false information that heads our way by taking a machine-learning approach to identify “fake news” on Twitter.
Ayton’s work as a research assistant at Pacific Northwest National Laboratory (PNNL) uses machine learning — a concept related to artificial intelligence — that uses text, associated images, mentions and hashtags of tweets and separates them into seven classes. These categories include tweets that are verified, propaganda, clickbait, satire, conspiracy, hoaxes, or have disinformation.
With the amount of unverified content coming through social media feeds and other online sources, Ayton’s research makes filtering out any fabricated content that much easier
“There’s just so much information out there it’s hard for yourself to keep track,” Ayton said. “Because you could have never heard of a source before so who knows what it’s going to be.”
Ayton said she and her research team developed five different models that take into account the different features of tweets. Depending on the combination of data, these models make predictions about which class the computer thinks the tweets should be labeled as.
“For the text-only model, we would turn each word into a number or a vector of numbers to represent that word, and then you take all of those vectors and multiply it with a matrix,” she said. “The machine learning part is determining those values inside the matrix so the resulting vector will give a probability for each class. The highest probability is the class the tweet belongs to — that’s what we call a neural network.”
The project deals with “deep” neural networks that involve layers of matrices with different functions. Her project uses a long short term memory (LSTM) layer to capture different aspects of the language used in tweets. The long and short term memory matrices help make sure the computer doesn’t lose any important information within the tweet.
Machine learning is a subset of artificial intelligence that uses statistical techniques to give computers the ability to ‘learn’ with data.
The project is part of a larger venture at PNNL that is more concerned with how users can interact with the technology, though Ayton has some ideas for its applications on Twitter.
“It could be applied to everyday situations, most likely as a tool for users or journalists for the purposes of fact-checking.” Ayton said. “There could be a flag that says, ‘This was predicted as verified or clickbait content’ or something like that, just so you know what you’re reading. But the user doesn’t have to trust it at all.”
Ayton said successfully detecting fake news on Twitter is still a work in progress, but the technology is better than just random guessing. She and her research team recently submitted a paper to the Advances in Social Networks Analysis and Mining (ASONAM) conference that includes all the project results and analyses.
Prior to her research assistantship, Ayton said she worked on several other projects — some successful, others not so much. In her first experience with machine learning, her research group analyzed New York Times articles and metadata in an attempt to remove bias, which later evolved into modeling the different writing styles of authors.
Still, Ayton wants to continue working on projects similar to this one.
“I want to work on projects that have a larger impact than just, ‘Oh look I did this cool thing,’” she said. “I want to do something that can be applied to the real world.”