Detecting statements most likely to ignite debate
Can artificial intelligence pinpoint the most contentious statement in a news article?
Artificial Intelligence (AI) is being put to all sorts of purposes. From diagnosing illness to recommending music.
The field of AI that deals with deriving insight from text is called Natural Language Processing (NLP). While the breakthroughs have been coming, it’s still early days for NLP. There are many more applications of NLP technology yet to be explored that will prove positive for individuals and society.
Aiding interpretation and improving engagement
Despite the printing press having been invented over 500 years ago, text still dominates as the most popular and powerful medium for quickly and accurately conveying ideas.
And yet understanding and interpreting text is a relatively rare and highly sought after skill.
In Australia for example, a wealthy First World country, almost half of all adults do not have the literacy competence required to meet the increasingly complex demands of a knowledge society, including connecting concepts together and critical thinking.
At Worldview Exchange we’re building the Debate Detector, a tool for identifying the most contentious statements within news and current affairs reporting as a potential point of ignition for private inquiry or public debate.
By sorting opinion from fact, the Debate Detector will enable any reader to more easily find and engage with the ideas that are most contentious and debatable, aiding understanding and provoking critical thinking.
Below is a short description of the NLP techniques we’ve employed.
How does the Debate Detector work?
There are two main challenges to address in detecting the most contentious statements as a potential point for igniting debate.
The first is filtering the news article for subjectivity. Contentious statements tend to be where the author, or someone being quoted, is going out on a limb, making a proposition or conveying a sentiment or feeling or belief, and not making a simple statement of fact.
Debate Detector uses machine learning to assess the degree of subjectivity present in each sentence within an article.
Subjectivity alone however is not enough to find the very best statement to start debate. There is often lots of subjectivity, particularly in opinion pieces, but not all of it is on topic.
So the second challenge is assigning a score to each sentence based on how well they capture the central thesis of the article as a whole.
Debate Detector groups the article with other articles on the same issue published online around the same time. It scans all articles and creates a map of topics, made of words most closely linked to each other within the articles. Debate Detector then rates each sentence in the original article according to how closely it relates to the dominant topic.
Finally, sentences are ranked as a product of both their subjectivity and relatedness to the central topic.
Results
A demo of the Debate Detector is now live. Try it out.
The Debate Detector has shown positive and consistent results early on with the promise of further improvements with more training.
It detects more lower ranked sentences in articles that focus on facts, like crime reporting, and finds more higher ranked sentences in opinion pieces. It demonstrates fidelity in distinguishing subjective from objective text within articles.
Potential purposes
The Debate Detector has a range of other applications beyond being used to improve active reader engagement in news consumption.
Journalists could use Debate Detector to review their own copy, to check for opinion or speculation in fact based reporting. Or to see where they’ve gone the furthest off topic.
Editors can deploy the tool for the same purposes.
