We are tremendously excited to announce today the official launch of the GDELT Live Trends Dashboard. Now when you visit GDELT Live, instead of seeing only a blank search box, you are now presented with a summary of worldwide activity as seen through the eyes of GDELT, updated every 15 minutes. At the top of the page is a Visual News Map display of major stories worldwide, followed by a list of the top 60 trending narratives worldwide that are rapidly increasing in volume, followed at the bottom by coverage deemed of potential relevance to currently declared UN OCHA/ReliefWeb Disasters.
To generate the Visual News Map at the top of the page, GDELT examines all monitored coverage over the preceding 45 minutes and calculates for every location on Earth a list of articles mentioning that location in order of relevance to that location, relevance to emerging themes at that location, and level of detail in the article. It ranks every location in order by the total volume of coverage about it during the most recent 45 minutes and adds each to the map in order, selecting a maximum of three locations per country to ensure broad coverage over the world. At each location, GDELT displays the social sharing image for the article it deems most representative of its coverage of that location during the last 45 minutes, with a link to the relevant article. Note that you may seen broken images on this map, and articles may be in any of the 65 languages live-translated by GDELT. Only news coverage that includes a social sharing image is represented on this map.
Note that images may appear in unusual locations on the Visual News Map for a variety of reasons. Attempting to properly identify and separate the relevant news article from the surrounding body container for news coverage from nearly every corner of the early across a myriad of technical platforms and 65 languages is extremely difficult, leading to cases where material is misidentified as part of the article that was in fact part of an unrelated inset box or coverage blurb. Machine translation is still far from perfect, and attempting to live-translate the world's news from 65 languages in realtime will inevitably yield more than a few errors in translation. Humans are enormously creative in how they specify location in textual form and massive multilingual geocoding across 65 languages is breathtakingly difficult, involving a host of issues like assumed locality, yielding a fair level of error. Even if there is no error in identification, translation, and geocoding, an article may still mention multiple locations in very different parts of the world and GDELT may not place the image in the right one of those locations.
The Trending Narratives section of the page attempts to understand narratives trending across the world. Similar to the Visual News Map, it compiles a spatial map of the entire planet, calculating a complete inventory for every location on earth of every article mentioning that place. It further breaks down each article into its component themes, creating a spatio-thematic grid over the world, computing the density of every theme in the coverage of every location. In essence, for even the most remote hilltop on earth, GDELT computes the relevant breakdown of all coverage of that hilltop into its component thematic foci. It then compares each location+theme pairing over the past 45 minutes with the preceding 45 minutes and the preceding 6 hours. For each location+theme, it performs basic time series analysis to determine its rate of change over both windows and ranks every location+theme pairing globally into a single master table in order of whether it is exhibiting burst, breakout, trending, or stabilizing behavior, and compiles a final list of the top 60 emerging narratives. Each narrative represents a theme that is increasing rapidly at a specific location in space. A narrative is NOT the same as a "story" – a trending narrative could be caused by a single story that is attracting substantial attention, or a confluence of unrelated stories, each of which is receiving only minor coverage, but when taken together, combine to represent a substantial increase for that location.
It is important to recognize the distinction between the "narrative" approach that GDELT uses and the "story" approach used by Google News and other systems. Google News performs basic keyword clustering on the coverage it monitors, grouping together sets of articles that share largely the same wording. This clusters multiple articles about a given situation into a single "story" that allows one to compare the similarities and differences in how the world's news outlets are covering that situation. In contrast, GDELT examines the world at the "narrative" level, looking across all stories referencing a given location to find the dominate themes that are prevailing across all of those stories. A good example of this might be a small town where an anti-corruption bill is being debated by the city council, the local mayor was arrested the month before on corruption charges, an NGO has halted funding to a local development project for corruption concerns, and there are protests in front of the courthouse about corruption. Each of these activities represent their own independent stories that story-centered approaches like that used by Google News would cluster as independent stories with no obvious connection among them. GDELT's narrative-centered approach would instead identify the town as having a trending narrative of "Corruption" that unifies the coverage emerging about it. This becomes critically important to understand the broader climate and context in which local actors are engaging. A reader coming across an isolated news article about any of the situations above would likely dismiss it as a one-off situation, but seeing the confluence of all of those stories together paints a far different picture of a local citizenry rising up against endemic corruption, which has considerable bearing on the local political environment and stability.
Finally, at the bottom of the page, GDELT lists any coverage it believes may be of possible relevance to active UN OCHA/ReliefWeb Disasters. Note that some of this coverage may suffer from the same false positives as above. In other cases, GDELT may determine that an article has relevance to a disaster due to the events it describes potentially impacting a disaster. For example, an article may describe an emerging pandemic in a neighboring country heading towards the border of a country with massive flooding. While this may not be directly related to the flooding, it provides context of potential further humanitarian impact with the risk of disease spread to an area that is highly vulnerable to outbreaks.
By default the GDELT Live homepage now displays the latest trending global information, updated every 15 minutes. You can also use the search box to display information on a specific leader, organization, location, or theme. Note that only trending stories will be displayed, and thus a story that is receiving considerable media attention, but whose coverage volume has reached peak and has not changed substantially over the past 45 minutes will not be displayed in these search results. Instead, a link allows you to switch to "Explorer View" which offers an enormous array of incredibly powerful analytic tools such as network diagrams, heatmaps, language breakdowns, etc.
We've based GDELT Live Trends on the incredible feedback we've been hearing from all of you, and we can't wait to see what you're able to do with this new capability!