Using Natural Language Processing to Combat Filter Bubbles and Fake News – 360° Stance Detection
One of the most alarming cultural developments in the internet age has been due to the way that we access news content. The creation of filter bubbles, where people only consume media that expresses views that they are likely to agree with, has led to entrenchment of existing biases in society.
This problem of filter bubbles is due in large part to the success of content recommendation systems on social networks, smartphones, and websites. The content recommended to the user is heavily oriented toward the same topics and news outlets that he or she has already shown an interest in. Because of this, the views expressed in the recommended content are likely to be similar to what the user has already read, making the content people consume resemble an echo chamber.
To make matters worse, filter bubbles also increase people’s susceptibility towards ‘fake news’, where fabricated news stories spread sensational but false information to readers with a particular stance on a social or political issue. In the final three months of the presidential election in 2016, the most popular fake news stories were shared even more than the most popular legitimate news stories on Faceboo
Facebook’s attempt to mitigate the problem by placing red flags next to content from known fake news sites has come to an unsuccessful end, and the Natural Language Processing community has been making efforts to ameliorate this problem.
360° Stance Detection
To contribute to these efforts, our Science and Engineering teams at AYLIEN have built 360° Stance Detection. This tool gathers multiple news stories about a topic that also mention an entity, and classifies the stance each author is expressing towards a given topic – ‘for,’ ‘against,’ or ‘neutral’.
This tool allows us to automatically source news stories that take different views about a certain topic. For example, we can search for stories about Brexit that mention Theresa May, and retrieve stories that the tool has recognized as supporting Theresa May as well as stories recognized as being against Theresa May.
360° Stance Detection Demo
To showcase how the tool works, we’ve put together a simple demo that you can access here. Given a topic and a keyword, the stance detection tool will gather stories about a given topic and predict the stance of the author of each story toward the keyword. The demo will then display the results on a scatter plot that graphs the stance of the author along with the popularity of the website (according to its Alexa ranking).
Below is a screenshot of the scatter plot showing the predicted stance of each author toward Donald Trump in stories about Robert Mueller. Mueller is the special counsel charged with investigating Russian meddling in the 2016 Presidential Election, a news topic that prompts polarized opinions about Trump, depending on your political stance.
You can see that the model has predicted the stance of a story about Mueller on alt-right stalwart Breitbart as being pro-Trump, while a story about Mueller appearing on Real Clear Politics a favourite of the left, has been predicted as being anti-Trump. We can also see the stance predicted about Trump in stories about Mueller by a diverse range of publishers like ABC, The Sun, Farming UK, and others.
Here’s a two-minute walkthrough of the demo by our founder, Parsa, that predicts the stance of authors toward Ireland in stories that mention both Ireland and Brexit:
In practice, this tool can be used to make it easy for people to access recommended content that does not simply agree with the content that they have already read, making it easier for people to regularly be exposed to see other points of view. It can also be used to analyze the stance of organizations and individuals toward a particular entity in the coverage of an event.