THE GDELT EVENT DATABASE
DATA FORMAT CODEBOOK V2.0
2/19/2015
http://gdeltproject.org/
INTRODUCTION
This codebook provides a quick overview of the fields in the GDELT Event file format and their
descriptions. GDELT Event records are stored in an expanded version of the dyadic CAMEO format,
capturing two actors and the action performed by Actor1 upon Actor2. A wide array of variables break
out the raw CAMEO actor codes into their respective fields to make it easier to interact with the data,
the Action codes are broken out into their hierarchy, the Goldstein ranking score is provided, a unique
array of georeferencing fields offer estimated landmark-centroid-level geographic positioning of both
actors and the location of the action, and a new “Mentions” table records the network trajectory of the
story of each event “in flight” through the global media system.
At present, only records from February 19, 2015 onwards are available in the GDELT 2.0 file format,
however in late Spring 2015 the entire historical backfile back to 1979 will be released in the GDELT 2.0
format. The Records are stored one per line, separated by a newline () and are tab-delimited (note
that files have a “.csv” extension, but are actually tab-delimited).
With the release of GDELT 2.0, the daily GDELT 1.0 Event files will still be generated each morning at
least through the end of Spring 2015 to enable existing applications to continue to function without
modification. Please note that at present, since GDELT 2.0 files are only available for events beginning
February 19, 2015, you will need to use GDELT 1.0 to examine longitudinal patterns (since it stretches
back to January 1, 1979) and use GDELT 2.0 moving forward for realtime events.
There are now two data tables created every 15 minutes for the GDELT Event dataset. The first is the
traditional Event table. This table is largely identical to the GDELT 1.0 format, but does have several
changes as noted below. In addition to the Event table there is now a new Mentions table that records
all mentions of each event. As an event is mentioned across multiple news reports, each of those
mentions is recorded in the Mentions table, along with several key indicators about that mention,
including the location within the article where the mention appeared (in the lead paragraph versus
being buried at the bottom) and the “confidence” of the algorithms in their identification of the event
from that specific news report. The Confidence measure is a new feature in GDELT 2.0 that makes it
possible to adjust the sensitivity of GDELT towards specific use cases. Those wishing to find the earliest
glimmers of breaking events or reports of very small-bore events that tend to only appear as part of
period “round up” reports, can use the entire event stream, while those wishing to find only the largest
events with strongly detailed descriptions, can filter the Event stream to find only those events with the
highest Confidence measures. This allows the GDELT Event stream to be dynamically filtered for each
individual use case (learn more about the Confidence measure below). It also makes it possible to
identify the “best” news report to return for a given event (filtering all mentions of an event for those
with the highest Confidence scores, most prominent positioning within the article, and/or in a specific
source language – such as Arabic coverage of a protest versus English coverage of that protest).
剩余10页未读，继续阅读
评论0