Automated attempts to identify problematic texts from their content include Google’s ‘hate speech AI’ and China’s keyword-based censorship of social media. Twitter attempts to detect bots with humans reporting.
Other attempts exist. For example, “Our work on the Credibility Coalition, an effort to develop web-wide standards around online-content credibility, and PATH, a project aimed at translating and surfacing scientific claims in new ways, is part of two efforts of many to think about data standards and information access across different platforms. The Trust Project, meanwhile, has developed a set of machine-readable trust indicators for news platforms; Hypothesis is a tool used by scientists and others to annotate content online; and Hoaxy visualizes the spread of claims online.”
However, these attempts can be fooled by manipulating the exact words used in an article (or tweet), and have issues with detecting sarcasm, irony, criticism of the problematic texts, and other subtle elements of discourse. For some mediums such as videos (e.g., beheadings by ISIS) or photos, text search obviously does not work and other methods are employed which are not satisfactory.
One notable attempt is Trustrank, which attempts to combat web spam by defining reliability. TrustRank uses a seed of reliable websites (selected manually) and propagates reliability by using Pagerank. Notably, TrustRank does not utilize passive data collected from user behaviors, or measures of user reliability.
We describe an automated, robust fake news detector which we call Human Interaction News Trust System [HINTS] to detect fake news and misinformation even in the presence of adversaries who know how the detector works. Our key tools are network dynamics and classification of members of the network in terms of their historical interaction with news. We look at how known and suspected fake news propagates in a dynamic network of people, and use this data to identify new posts/publication/news items that are likely to be fake as well. This also gives us information about accounts controlled by an adversary. Platforms can use this data to limit the damage a fake news article can do, by limiting the reach of such an article. And while limiting its reach, they can still increase confidence in the fakeness of the article e.g., by making it visible to small groups of users whose use patterns are the strongest indicators.
Our solution works for a wide variety of classification problems. Below, you see our solution graphically represented. We used tweets with the hashtag "#Kashoggi" as our seeds, and we were able to detect fake news about Beto O'Rourke, the 2020 presendential hopeful.