[ddj] new api for mapping the geography of global news coverage

kalev leetaru kalev.leetaru5 at gmail.com
Wed Apr 26 16:53:27 UTC 2017

Apologies for cross-posting - thought many on this list might find of great
interest and utility the new GDELT mapping API for creating point, ADM1 and
country-level maps of the geography of global news coverage from nearly all
countries worldwide in 65 languages over the last 24 hours, updating every
15 minutes. The API, which is fully free and open, generates both instant
embeddable browser-based interactive maps and GeoJSON optimized for use
with platforms like Carto.


Specify any keyword or phrase and search the English machine translations
of all content monitored in those 65 languages, allowing you to search
across languages. For each keyword, the system compiles a list of all
locations (down to the resolution of a hilltop in many areas) that were
found within a sentence or two of your keyword and constructs a map showing
the locations mentioned most frequently in context with your search. You
can also map specific languages, domains, by tone, etc.

Perhaps most uniquely, we are releasing a set of experimental maps that
apply the deep learning image categorization we perform on global news
imagery each day (more than a quarter billion images processed last year)
and let you search by 10,000 labels of objects and activities depicted in
the image, the OCR'd text in more than 80 languages depicted in the image,
all of the text contained in the image file's metadata fields, the textual
caption of the image as it appeared in each article, and the result of a
Google Images reverse search that compiles a list of all of the captions
used for that image anywhere it was found on the open web and assigns
several million topical labels.

As but one simple, but extremely powerful example - one of our research
threads revolves around how violence is depicted across the world and the
differing levels of normative baselines (for example, here in the US ISIS
beheadings are typically shown with a "before" image or a heavily pixelated
image, whereas in the presses of certain other countries the raw graphic
image is frequently shown; similarly in the US we rarely see imagery of
drowned refugees with the notable exception of Alan Kurdi, while the
presses of other countries run graphic imagery of those who perish on a
more frequent basis). Understanding how violence is depicted in the presses
of the world and how those baselines are changing offers a lot of insight
into the question of desensitization and the communication of crises.

The new API allows you to create such a map in just a few seconds and have
it live update every 15 minutes - here is one such example map that
displays up to five images from the domestic press of each country over the
last 24 hours that were determined to potentially depict some sense of
"violence" (click on each country to see the images from its local press).
(WARNING: many of these images are very disturbing). While you will see
some errors here and there, overall this gives a very visceral sense of the
differences in depiction of violence throughout the world.


Similarly, here is a map of rubble, destruction, flooding and fire that we
are using in a series of forthcoming projects to ground truth the severity
of natural disasters as they occur in realtime (note that this particular
map below shows the imagery of damage FROM the press of that country, which
may reflect events in other countries as well):


Looking at text, you can map a particular news outlet like AllAfrica:


Or source language like Chinese:


Or the phrase "Donald Trump", aggregated to the country level:


You can find many more examples, along with full documentation on the
announcement this morning:


Feel free to email me directly with any questions! We are super excited to
see what you all are able to do with these new capabilities! And stay tuned
for our temporal API being released in a few weeks!

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.okfn.org/pipermail/data-driven-journalism/attachments/20170426/42d68e1b/attachment-0002.html>

More information about the data-driven-journalism mailing list