What is Trustium?

It is no secret the internet is full of fake news, misleading articles, clickbait headlines, and other misinformation. According to a study conducted at Dartmouth University, roughly one in four Americans visited a fake news website over a five-week period. Often articles are completely crafted fiction, resulting in mass consumption of misleading or false information. The spread of outright lies leads to even more polarization of misinformed citizens, causing everything from family disputes to swayed elections.

To make the situation even worse, research by the Pew Research Center shows that nearly 50% of people who share an article do so without reading the text, but share based on the sensational headline alone. Since online outlets have become more and more dependent on “Clickbait” headlines, they often don’t even match the stories they link to! A study by venngage.com found that 79% of the articles had an element of shock in the title. 17% were “listicles” and 29% had “you,” “I,” or hinted to a personal story in the title. This phenomenon is purely due to the advertising business model driving these online sites. The need to attract a page load requires them to do anything they can to get a click, and these tricks get people to click even if the content has no relation to the headline.  

Social Media continues to make these problems worse. Users are presented with “news” and other articles that the platforms want you to see, primarily to increase clicks, and thus revenue, for the company. Increasingly, this means showing only one side of any issue and posts from people who agree with you already. Some refer to this phenomenon as the “echo chamber,” where you will will only see posts that agree with your current ideas and beliefs on issues, and additional preferences in everything from colors to brands. This “echo chamber” effect tends to divide the population along ideological lines. As a response to pressure resulting from controversy in social media, the implemented “solution” has not been to notify users about potentially misleading content, but rather to present it lower in their newsfeed.

What if the end user knew whether to actually trust the article even before opening the link? What if each article had a badge that indicated whether it was a trustworthy article, whether the headline matched the content, and whether the source frequently generated hoaxes?  What effect would that kind of instant vetting of articles have on individuals, communities and society overall?

Trustium’s Vision 

Trustium helps people sort through it all and determine what is likely to be real news rather than sensationalism, hoax, or even just biased reporting.

Trustium employs a panel of experts, a Machine Learning Model, and a wider community to categorize articles.  Every link a user visits is evaluated to determine if it’s presenting a news story, and whether that story is biased or not.  In addition, Trustium is able to identify satire, opinion pieces, analysis, and entertainment stories and let the user know which bucket the page falls in.. This categorization is based on the source, writing style, content, references to other links and sites, and a variety of other criteria.  

The categorization is done in real time, so the reader knows immediately how trustworthy the information they are digesting is.

Experts Panel

The Experts Panel is composed of Journalists and other highly trusted individuals.  This panel will help to verify the training and verification data for the Machine Learning model. Our team of experts and respected journalists from a wide spectrum of viewpoints ensure we identify bias from fact, and that information is analyzed based on rhetoric and objective evidence rather than subjective opinions on issues.

Machine Learning Model

The Machine Learning model was created using the training and verification data from the Experts Panel. The system first creates a mathematical model (based on it’s rhetorical devices and structure) for each article as it is loaded, and sends that to the Machine Learning model for evaluation.  The model responds back immediately with probabilities detailing how likely that particular article is to be real news.

We don’t try to tell you what’s true, instead we give you a good basis for knowing whether what you’re reading is likely to be a well-written journalistic piece that is unbiased and presents a real news story.