Assessing trustworthiness of online content using community annotation and natural language processing
Millions of pieces of content created every day on the internet may be biased, hateful or misleading. The rise of negative content, fake news and misinformation is increasingly recognised in politics, but its reach is much wider; it can infiltrate trusted news sources, as well as smaller platforms like lifestyle blogs. While the speed and spread of information exchange can have many positive effects, it also poses a challenge to verification. It is difficult for humans alone to carefully analyse the vast amount of content on social media and news sources in order to correctly identify what information is spurious and what is trustworthy.
The AI technology developed by Factmata uses natural language processing (NLP) methods to automatically detect and categorise online content that might be harmful or untrustworthy. They have developed a dedicated annotation platform for content and recruited 24 specialist communities and over 2,000 experts – including journalists and researchers – to actively generate a training dataset. Training by the Factmata communities is ongoing to ensure that the algorithm continuously improves and is able to accurately categorise new types of content. The AI assigns a trust rating to all content that it examines, acting as a triage system for deceptive content. The trust rating assesses reliability from the perspective of nine different signals, including toxicity, political bias, clickbait and others to give a comprehensive overview of quality, safety and credibility. Factmata is in the early stages of developing this technology, but ultimately, they aim to develop a tool that individuals and companies can use to verify claims across any type of online content.
It remains to be seen how effective Factmata’s AI solution will be when the company reaches full maturity. Their original technology was developed for the verification of claims about statistics and it is unclear how well it adapts to more qualitative and nuanced statements. Even the most sophisticated NLP techniques are far from the contextual understanding and emotional intelligence used by humans in their interpretation of written content. Factmata hopes to overcome these barriers by involving communities and experts in the assessment of statements automatically detected by AI. In the meantime, they have made their technology available through an API, which accepts any URL link from the internet and returns an assessment of the content, based on the signals that they track.