(Photo/stdaily.com)

The “rumor shredder”, an algorithm model that can identify fake news, was recently created by scientists from the Alibaba DAMO Academy, an institute dedicated to exploring the unknown through scientific and technological research and innovation under China's e-commerce giant Alibaba.

Such an invention has the potential to remedy the widespread phenomenon on Chinese social media of fake news and pseudoscience reposts.

Li Quanzhi, leader of the R&D team of the algorithm model, explained that the “rumor shredder” could find original information and analyze whether the publisher is “reliable” based on their profession and previous posts.

Then, it would seek all available pieces of relevant information online to find out if they were published by authoritative organizations. Finally, the algorithm model would extract key arguments from the data and contrast with authoritative knowledge bases.

"The ‘rumor shredder’ is based on artificial intelligence (AI), though it does sometimes think like a human,” Li noted.

He said that the AI technology used can do things beyond human capabilities. For instance, when a piece of fake information is spreading fast on the internet, it is difficult for human beings to identify its authenticity as they don’t know who has received the information or why people repost it. However, that is not a problem for AI.

AI is able to contrast between the piece of information and existing knowledge bases very quickly. “People can make a basic judgment, but lack such knowledge bases,” he said.

The training of the “rumor shredder” is a complicated and time-consuming process, Li explained, adding that it called for 200 million pieces of information and millions of pieces of news to solidify language samples.

"We can’t say it’s 100-percent accurate, but we can identify whether a piece of information is a rumor,” Li noted.

In the just-concluded SemEval, a competition for evaluations of computational semantic analysis systems, the “rumor shredder” created a new record regarding fake news identification with an accuracy of 81 percent.

It is possible for such technologies to be applied in more scenarios. For instance, it would be able to detect plagiarism in an academic thesis by identifying writing styles, methodologies and themes. Additionally, the technology could help police spot the authors of rumors.

"Of course, this algorithm model still needs more language training, as it is currently unable to understand ironies and metaphors,” Li said, adding that he and his team would continue their research on the technology.