As society searches for new approaches to combating the misinformation, disinformation, digital falsehoods and foreign influence that plague the modern Web, one intriguing idea is that of creating coverage maps that capture where a claim first appeared, its spread through the informational ecosystem and the places that are covering it most. Such maps could go a long way towards lending critical additional context to contentious debates.
What might it look like to think of claims in terms of where they are resonating rather than the arguments they make or the sources they cite?
Imagine an elected official whose claims of success appear only on their own Website and a smattering of party sites, while all other sources refute those claims.
While such coverage skew would not necessarily prove that those claims were suspect, it would certainly raise concerns that they should be treated more cautiously than those covered widely across independent Websites, reference outlets and news reporting.
A diehard supporter of that candidate might not be swayed that there is any concern with their claims and might in fact believe those claims even more in the face of such evidence. The rest of society, however, would be able to more readily distinguish between widely supported claims and those argued only by the candidate themselves.
Similarly, a claim that appears only in news outlets traditionally associated with one political party, while those supportive of the other party make no mention could raise concerns of partisanship. A claim appearing on Democratic-leaning Websites that Obama was the greatest president or a claim appearing on Republican-leaning Websites that Trump was the greatest president lies entirely in the eye of the beholder, but by capturing the partisan divide in the coverage of those claims a reader can be more aware that the topic is disputed.
Of course, few readers would need a machine to tell them that the question of who was the “greatest” president or which party’s policies are the “best” might exhibit a partisan divide.
Instead, the power of such techniques comes from their ability to statistically identify and quantify topics that exhibit coverage divides.
For example, the open data GDELT Project’s Television Explorer, which searches the Internet Archive’s Television News Archive, is frequently used to quantify the degree to which a topic appears more on one television news station than others, quantifying particularly contentious topics and debates.
How might such a system work in practice?
Automated anomaly detection tools like Google’s Cloud Inference API could be used to automatically sift through coverage from different outlets to discover topics and language more associated with one set of outlets than others. Television news, with its fixed set of stations and well-understood political leanings makes such tasks particularly straightforward.
Online news can similarly be forged into similarity clusters based on related coverage or the billions upon billions of hyperlinks news articles provide to the outside world. Connecting news outlets based on how often they link to each other can provide critical context.
In fact, forming the billions of inter-outlet links of online news coverage into a single massive network diagram of the world’s news outlets offers a “you are here” media map that can be used to understand a given outlet’s position in the global news ecosystem, showing whether it is mostly linked to by a small cluster of partisan outlets or whether it is a widely reference source that crosses the partisan divide.
Such a media map could easily be used to automatically assess the partisanship and coverage clustering of a given story.
To assess the partisanship of a given story, such a system would compile a list of all of the outlets covering that story. The position of each of those outlets in the global media map and their respective cluster membership would be examined. Stories covered by outlets belonging only to a handful of small clusters could be identified as narrow interest, while those covered by outlets in major clusters or whose covering outlets span many clusters could be identified as more general interest stories.
Looking across large numbers of political stories, the “partisanship” of each cluster could be identified in terms of whether stories covered by outlets in that cluster are frequently covered by outlets in other clusters as well or whether they tend to be unique to that cluster.
The end result would be a score for each news story as it cascades across the informational ecosystem that assesses whether its debut and trajectory are primarily within narrow partisan clusters or whether it appears to transcend divides and is covered across the entire spectrum.
One could imagine a global network diagram of the media updating in real-time to show the story’s spread.
While such a system would not provide insight into the veracity of a story’s claims, it would provide critical context into its partisanship, identifying stories where partisan agenda setting is helping to drive its spread.
The system would not catch stories that are widely covered from opposing perspectives. A topic that is covered widely in one partisan sphere might generate a widespread reaction on the other side dismissing it, which would lead a coverage map to show it spanning the political divide. This could be addressed through enhancements tracking whether each outlet’s coverage was supportive or refutative.
Of course, such coverage maps need not track only news outlets. Such maps could be similarly used to map the spread of stories across social platforms. Even more powerfully, they could track the two mediums simultaneously, showing how mainstream and social media interact to advance or obstruct a story’s spread.
They could also flag geographic and demographic discrepancies in story spread, rather than assessing the partisan divide, offering even richer opportunities for understanding the divides that pull at the threads of the societal fabric.
Putting this all together, creating coverage maps of claims in real-time as they debut and spread across the informational landscape could go a long way towards helping lend critical context to viral memes, helping flag disputed claims and identify bipartisan coverage.
