gdeltr2

gdeltr2 — R’s modern GDELT Project interface

What is the GDELT Project?

The Global Database of Events, Language, and Tone [GDELT] is a non profit whose initiative is to:

construct a catalog of human societal-scale behavior and beliefs across all countries of the world, connecting every person, organization, location, count, theme, news source, and event across the planet into a single massive network that captures what’s happening around the world, what its context is and who’s involved, and how the world is feeling about it, every single day.

GDELT was founded in 1994 and it’s data commences in 1979. Over the last two years the GDELT’s functionality and abilities have grown exponentially, for example in May 2014 GDELT processed 3,928,926 where as in May 2016 it processed 6,198,461. GDELT continues to evolve and integrate advanced machine learning tools including Google Cloud Vision, a data store that became available in February 2016.

This package wraps GDELT’s four primary data stores

• The GDELT Events Database [EVENTS]: Global Events, 1979 to present.
• The GDELT Global Knowledge Graph [GKG] : GDELT’s Knowledge Graph, April 2013 to present.
• The GDELT Full Text API [Full Text API]: Full text search for all monitored sources within a 24 hour window. Output includes raw data, sentiment, and word counts.
• The GDELT Visual Knowledge Graph VGKG: Google Cloud Vision API output for every indexed piece of GKG media.

Why gdeltr2?

My main motivation for this building package is simple, GDELT IS INCREDIBLE!!

Accessing GDELT’s data gold is doable but either difficult or costly.

Currently, anyone proficient in SQL can access the data via Google Big Query. The problem is that even if you want to use SQL, users have to pay above a certain API call threshold and then you still need another layer of connectivity to explore the data in R.

Although R has two existing packages that allow users to interact with portions of GDELT’s data outside of Big Query:

• gdeltr
• GDELTtools

These packages are old, incomplete and difficult to use. It is my hope that gdelt2r allows the R user easy access to GDELT’s data allowing for faster, more exhilarating data visualizations and analysis!

PRIOR TO INSTALL

This package may require the development versions of devtools and dplyr so, to be safe, before installation run the following code:

devtools::install_github("hadley/devtools")
devtools::install_github("hadley/dplyr")
devtools::install_github("hafen/trelliscopejs")

Installation

devtools::install_github("abresler/gdeltr2")

Function Ontology

The package currently consists of two function families, data acquisition and data tidying.

The package’s data acquisition functions begin with get_urls_ for acquiring data store log information, get_codes_ for acquiring code books and get_data_ for downloading and reading data.

The data tidying functions begin with parse_ and they apply to a number of the features in the gkg and vgkg data stores that will get described in further detail farther below.

CAUTION

• gdeltr2 requires an internet connection for any data retrieval function
• The package’s get_gkg_data and get_gdelt_event_ functions are extremely bandwidth intensive given the download sizes of these data stores.
• The package is very memory intensive given the unzipped size of the GDELT Event, Global Knowledge Graph and Visual Knowledge Graph files.

Primary Functions

• Full Text API
  • ft_v2_api() - retrieves descriptive data from V2 API see this blog post for more on how to use this
  • ft_trending_terms() - retrieves trending terms over the last 15 minutes. The term can be a GDELT tag, location, person, place, or thing.
• GDELT Events
  • get_urls_gdelt_event_log() - retrieves descriptive data and urls for all available GDELT event downloads.
  • get_data_gdelt_period_event_totals() - retrieves summary event data for a given a period [monthly, daily, yearly]; this can be grouped by country.
  • get_data_gdelt_periods_event() - retrieves GDELT event data for a specified periods. Periods are by 4 digit years from 1979 to 2005, 6 digit year month from January 2006 to March 2013, and 8 digit year month day code thereafter.
• Global Knowledge Graph
  • get_urls_gkg_15_minute_log - retrieves GKG 15 minute capture logs; data begins February 18th, 2015 for the three table types
    • gkg: This is the full gkg data set and contains columns that may require further data tidying tying to a GKG Record ID
    • export: This data replicates the output contained in the GDELT event table for processed documents tying to a Global Event ID
    • mentions: This data contains information surrounding the processed events, including sources, tone, location within a document and this tying to a Global Event ID
  • get_urls_gkg_daily_summaries - retrieves daily gkg capture logs; data begins in April of 2013.
    • Each day contains a count file and the full gkg output.
  • get_data_gkg_days_summary() retrieves GKG daily summary data for specified date(s), this captures count files by is_count_file = T
  • get_data_gkg_days_detailed() - retrieves GKG data from the data cached every 15 minutes for specified date(s) for a given table. The table can be one of c('gkg', 'export', 'mentions'). This function may require significant bandwidth and memory given the potential file sizes.
• American Television Knowledge Graph
  • get_urls_gkg_tv_daily_summaries() - retrieves available dates
    • gkg_tv_days() - retrieves data for specified dates. Note that the data is on a 2 day lag so the most recent data is 2 days old.
• Location Sentiment API
  • dictionary_stability_locations() - retrieves possible locations
  • dictionary_stability_locations() - retrieves instability data for a specified location and time period. Variables can be c('instability', 'conflict', 'protest', 'tone', 'relative mentions') Time periods can be c('daily', '15 minutes'), for daily the data is the average per day of the specified variable for the last 180 days and for 15 minutes the data is the variable reading every 15 minutes for the last week.
• Visual Global Knowledge Graph
  • get_urls_vgkg() - retrieves VGKG log urls
  • get_data_vgkg_dates() - retrieves VGKG data from the data cached every 15 minutes for specified date(s).

Tidying Functions

Many of the columns in the GKG output are concatenated and require further parsing for proper analysis. These function tidy those concatenated columns, note given file sizes the functions may be time consuming.

V2 Full Text API

You can refer to this blog post that discusses how to use this functionality.

Global Knowledge Graph

• parse_gkg_mentioned_names() - parses mentioned names
• parse_gkg_mentioned_people() - parses mentioned people
• parse_gkg_mentioned_organizations() - parses mentioned organizations
• parse_gkg_mentioned_numerics() - parses mentioned numeric figures
• parse_gkg_mentioned_themes() - parses mentioned themes, ties to CAMEO Theme Codes
• parse_gkg_mentioned_gcams() - parses resolved GCAMs ties GCAM code book.
• parse_gkg_mentioned_dates() - parses mentioned dates according to the GKG scheme
• parse_xml_extras() - parses XML metadata from GKG table

Visual Global Knowledge Graph

• parse_vgkg_labels() - parses and labels learned items
• parse_vgkg_landmarks() - parses and geocodes learned landmarks
• parse_vgkg_logos() - parses learned logos
• parse_vgkg_safe_search() - parses safe search likelihoods
• parse_vgkg_faces() - parses learned faces
• parse_vgkg_ocr() - parses OCR’d items
• parse_vgkg_languages() - parses languages

Code Books

All these the GDELT and GKG datasets contain a whole host of codes that need resolution to be human readable. The package contains easy access to these code books to allow for that resolution. These functions provide access to the code books:

• get_codes_gcam() - retrieves Global Content Analysis Measurement [GCAM] codes
• get_codes_cameo_country() - retrieves Conflict and Mediation Event Observations [CAMEO] country codes
• get_codes_cameo_ethnic() - retrieves cameo ethnic codes
• get_codes_cameo_events() - retrieves cameo event codes
• get_codes_gkg_themes() - retrieves gkg theme codes
• get_codes_cameo_type() - retrieves cameo type codes
• get_codes_cameo_religion() - retrieves cameo religion codes
• get_codes_cameo_known_groups() - retrieves cameo known group codes

Coming Soon

• Vignettes
• Generic data visualization functions
• Generic machine learning and data analysis functions
• bigrquery integration
• Third party database mirror

EXAMPLES

library(gdeltr2)
load_needed_packages(c('dplyr', 'magrittr'))

GDELT Event Data

events_1989 <-
  get_data_gdelt_periods_event(
    periods = 1989,
    return_message = T
  )

GKG Data

gkg_summary_count_may_15_16_2014 <-
  get_data_gkg_days_summary(
    dates = c('2014-05-15', '2014-05-16'),
    is_count_file = T,
    return_message = T
  )

gkg_full_june_2_2016 <-
  get_data_gkg_days_detailed(
    dates = c("2016-06-02"),
    table_name = 'gkg',
    return_message = T
  )

gkg_mentions_may_12_2016 <-
  get_data_gkg_days_detailed(
    dates = c("2016-05-12"),
    table_name = 'mentions',
    return_message = T
  )

GKG Television Data

gkg_tv_test <- 
  get_data_gkg_tv_days(dates = c("2016-06-17", "2016-06-16"))

GKG Tidying

load_needed_packages(c('magrittr'))

gkg_test <- 
  get_data_gkg_days_detailed(only_most_recent = T, table_name = 'gkg')

gkg_sample_df <- 
  gkg_test %>% 
  sample_n(1000)

xml_extra_df <- 
  gkg_sample_df %>% 
  parse_gkg_xml_extras(filter_na = T, return_wide = F)

article_tone <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_article_tone(filter_na = T, return_wide = T)

gkg_dates <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_dates(filter_na = T, return_wide = T)

gkg_gcams <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_gcams(filter_na = T, return_wide = T)

gkg_event_counts <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_event_counts(filter_na = T, return_wide = T)

gkg_locations <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_locations(filter_na = T, return_wide = T)

gkg_names <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_names(filter_na = T, return_wide = T)

gkg_themes <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_themes(theme_column = 'charLoc',
                                      filter_na = T, return_wide = T)

gkg_numerics <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_numerics(filter_na = T, return_wide = T)

gkg_orgs <-
  gkg_sample_df %>% 
  parse_gkg_mentioned_organizations(organization_column = 'charLoc', 
                                             filter_na = T, return_wide = T)

gkg_quotes <-
  gkg_sample_df %>% 
  parse_gkg_mentioned_quotes(filter_na = T, return_wide = T)

gkg_people <- 
  gkg_sample_df %>% 
  parse_gkg_mentioned_people(people_column = 'charLoc', filter_na = T, return_wide = T)

VGKG Tidying

vgkg_test <- 
  get_data_vgkg_dates(only_most_recent = T)

vgkg_sample <- 
  vgkg_test %>% 
  sample_n(1000)

vgkg_labels <- 
  vgkg_sample %>% 
  parse_vgkg_labels(return_wide = T)

faces_test <- 
  vgkg_sample %>% 
  parse_vgkg_faces(return_wide = T)

landmarks_test <- 
  vgkg_sample %>% 
  parse_vgkg_landmarks(return_wide = F)

logos_test <- 
  vgkg_sample %>% 
  parse_vgkg_logos(return_wide = T)

ocr_test <- 
  vgkg_sample %>% 
  parse_vgkg_ocr(return_wide = F)

search_test <- 
  vgkg_sample %>% 
  parse_vgkg_safe_search(return_wide = F)

Sentiment API

location_codes <-
  dictionary_stability_locations()
location_test <-
  instability_api_locations(
    location_ids = c("US", "IS", "CA", "TU", "CH", "UK", "IR"),
    use_multi_locations = c(T, F),
    variable_names = c('instability', 'tone', 'protest', 'conflict'),
    time_periods = c('daily'),
    nest_data = F,
    days_moving_average = NA,
    return_wide = T,
    return_message = T
  )

location_test %>%
  dplyr::filter(codeLocation %>% is.na()) %>%
  group_by(nameLocation) %>%
  summarise_at(.vars = c('instability', 'tone', 'protest', 'conflict'),
               funs(mean)) %>%
  arrange(desc(instability))location_codes <-
  dictionary_stability_locations()
location_test <-
  instability_api_locations(
    location_ids = c("US", "IS", "CA", "TU", "CH", "UK", "IR"),
    use_multi_locations = c(T, F),
    variable_names = c('instability', 'tone', 'protest', 'conflict'),
    time_periods = c('daily'),
    nest_data = F,
    days_moving_average = NA,
    return_wide = T,
    return_message = T
  )

location_test %>%
  dplyr::filter(codeLocation %>% is.na()) %>%
  group_by(nameLocation) %>%
  summarise_at(.vars = c('instability', 'tone', 'protest', 'conflict'),
               funs(mean)) %>%
  arrange(desc(instability))