Taiwan's fact-checkers are using AI to combat misinformation

Using artificial intelligence (AI) to scale up fact-checking is happening in Taiwan. The AI tool, Rumor Catcher, which has just been launched this summer, enables Taiwanese Fact-checkers to identify the most viral and harmful claims circulating on social media platforms and stop the spread effectively.

Taiwan FactCheck Center (TFC) has been teaming together with Taiwan Institute for Information Industry (III), developing AI tools since March 2020. Thanks to the technical support from III, the progress has been inspiring and promising.

The void in Chinese claims detecting

It has been a long journey- using technology to fight against misinformation has been a continuing research for fact-checkers worldwide since almost a decade ago. Among various impressive projects, the collective project of British Full Fact, South Africa's Africa Check, and Chequeado from Argentina, which receive funding and support from the Google AI program, has shown great hopes.

Their tools have been already used on a daily basis: Full Fact has detected over 500,000 claims made about COVID-19 in the UK media alone since the outbreak. In Argentina, half of the claims fact-checked each week are identified by AI. A big hurrah for our colleagues!

However, these achievements and hopeful tools are programmed for the English language, at most localized into Spanish. Overall, those abundant research energy and resources have not yet been extended to identify claims in the Chinese language.

As the first International Fact-Checking Network (IFCN) verified fact-checking organization in the Chinese speaking world, TFC's effort has influenced not only the audience in Taiwan, but also in Hong Kong, Malaysia, South-eastern Asia, and the Chinese communities in America and Canada. TFC’s editor in chief, Summer Chen (陳慧敏), has had a dream introducing technology into work.

“If artificial intelligence is able to generate fake news, can we also use AI to combat fake news? ” She asked.

To scale our work and keep pace with fact-checkers across the world, TFC and III have been developing AI tools specializing in monitoring claims from the Chinese context since March. We are very excited to announce that the first stage of this project, the AI filter tool, has gone online lately. This new tool has significantly changed TFC's workflow.

Breakthrough: Crowd Detecting, Real-Time Collecting, Automated Clustering 

Led by Bing-Yi Chen (陳棅易), the expert from Digital Service Innovation Institute of III, along with the support from Digital Transformation Institute and Cybersecurity Technology Institute, the first stage of the project "Rumor Catcher" has begun the trial run this mid-August. TFC and other Taiwan fact-checkering outlets have been since using this tool in daily work to improve and scale fact-checking efforts.

The AI tool compiles complaints from users, clusters similar claims in the same group, and gathers the statistics. Once we have the sets of similar claims, the machine learning method then checks if these claim sets match previous fact-checks- something is already known to be false, so fact-checkers don't need to recheck them.

Based on the statistics of how often one claim is reported, fact-checkers know the most viral claims circulating online. Fact-checking is time-consuming, so it is crucial to relocate the limited time and resources on debunking the most harmful misinformation. The tool can efficiently filter the volume of massive data, empowering fact-checkers to focus on the most urgent things to be fact-checking each day. 

Since the AI filter tool's debut, the data has been collecting and reporting to III as feedback to improve the AI tool. The meantime, III and TFC are starting the second stage of the project, the AI detecting tool.

Bing-Yi Chen's team started by determining and annotating elements that indicate misinformation, including the content, context, and other social traits of misinformation. After the massive study of international research and practice from Taiwan fact-checking organizations, Chen's team defined three advanced indicators to analyze the language feature, dissemination pattern, and publisher accounts of claims to spot misinformation, using artificial neural networks and deep learning methods. 

The AI tools, which is originally designed for fact-checkers to rapidly determine the reliability of a claim, is also applied in fields such as marketing and information communication business, says Chen. By the functions of filtering out unauthentic content and picking up influential information, the AI tools could serve social media analysis and system integration service firms well.

Fact-Checker's Trump Card

"By using the AI Filter tool to collect claims from multiple channels, our fact-checkers are empowered to monitor the development of claims as close as to real-time with objective data." Summer Chen said.

Embracing the new era, TFC now needs to adapt itself to the new tools in terms of adjusting the workflow. 

"The AI tools help dramatically increase the speed, reach, and impact of fact-checking in Taiwan, and push TFC one step further to the digital transformation." Said Summer Chen, whose dream is coming true step by step.

Taiwan FactCheck Center (TFC) is Taiwan's first independent nonprofit fact-checking organization, jointly founded by the Association for Quality Journalism and Taiwan Media Watch Foundation in 2018 to promote reliable information, digital literacy, and improve the information ecology in Taiwan. A Verified signatory of the IFCN code of principles.

Image: Unsplash/John Schnobrich