ML Enlisted to Slow the Spread of Fake News
As should be obvious by now, traditional fact-checking methods have failed to stop “fake news” from spreading like wildfire via social media circuits.
While platforms like Facebook (NASDAQ: FB) are developing new algorithms for ferreting out fake new that woulod supplement a planned army of human moderators, others are taking more pro-active approaches. Among them is MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) that is working with a Middle Eastern partner to shift the focus from fact-checking individual claims to the source of fake news.
CSAIL and the Qatar Computing Research Institute have come up with a machine learning tool used to determine if a news source is accurate or harbors a political agenda. “If a web site has published fake news before, there’s a good chance they’ll do it again,” said Ramy Baly, an MIT researcher and lead author on a paper about the fake news detector. “By automatically scraping data about these sites, the hope is that our system can help figure out which ones are likely to do it in the first place.”
The researchers said their scheme requires only about 150 articles to sniff out fake news and political agendas. The hope is that a system that can be trained more quickly than humans can fact-check assertions could stanch the flow of fake news by uncovering suspicious web sites.
The researchers trained their algorithm using data from the web site Media Bias/Fact Check (MBFC) that uses human fact-checkers to analyze more than 2,000 web sites ranging from mainstream news outlets to murky “content farms.” The machine learning algorithm uses the MBFC’s criteria for classifying web sites according to accuracy and political leanings.
The trained algorithm delivered 65-percent accuracy in spotting fake news from other sources and was about 70-percent accurate in detecting the source’s political slant.
Based on those results, the researchers concluded that the most reliable way to detect fake news and biased reporting were to examine “common linguistic features” in news stories that might supply clues to the author’s sentiment. The most common are the use of hyperbole or emotion—what has come to be typified by social media “rants.”
While the fake new detector has thus far yielded promising results in the effort to stop the viral spread of phony news reports, the research noted that the system would still have to work with human fact-checkers. In that respect, fact checkers could consult the scores generated by the machine learning algorithm in order to zero-in on specific sites spreading conjecture and political agendas rather than fact.
The project also generated an open-source data set of more than 1,000 new sources annotated with “factuality” and bias scores.