Analytics/Data Lake/Edits/Geoeditors

From Wikitech
Jump to navigation Jump to search

This is a private dataset that shows, in aggregate, what countries editors edit from. It is available only on the private cluster for now as we work out how to publish it for broader use. You can access this dataset on Superset or in the wmf.geoeditors_monthly Hive table. A public version of this data exists.

For details on the infrastructure that produces this data source, see Analytics/Systems/Geoeditors.

Aggregated data

There are three datasets available with aggregated data:

  • wmf.geoeditors_monthly counts editors by country and by wiki. Further, we publish two versions of this data: a yearly aggregate described below and a public bucketed version in the form of files that are updated every month.
  • geoeditors_edits_monthlycounts edits by country and by wiki
  • unique_editors_by_country_monthly counts editors by country regardless of wiki.

Geoeditors Monthly

The wmf.geoeditors_monthly table is defined as:

column data type description
wiki_db string The wiki database the editors worked in
country_code string The 2-letter ISO country code this group of editors geolocated to, including Unknown (--)
users_are_anonymous boolean Whether or not this group of editors edited anonymously
activity_level string How many edits this group of editors performed

possible values: "1 to 4", "5 to 99", "100 or more"

distinct_editors bigint Number of editors meeting this activity level
namespace_zero_distinct_editors bigint Number of editors meeting this activity level, with only namespace zero edits
month string [partition] The month in YYYY-MM format

NOTE: while users are grouped by their user_name[1], which is unique across all wikis, anonymous users are grouped by a hash of their User Agent and IP. This can still group multiple editors as one, but it should be rare enough that the numbers are indicative of anonymous editing patterns. Bots are identified as well as possible and filtered out, so this data does not include any bot activity we can identify.

From this table, you can find the number of editors, with different activity levels, anonymous or logged-in, in a specific project, editing from a specific country. An example query gets all this data for French Wikipedia in March 2018:

select 
    country_code, 
    users_are_anonymous, 
    activity_level, 
    sum(distinct_editors) as editor_count 
from wmf.geoeditors_monthly 
where month='2018-03' and wiki_db = 'frwiki' 
group by country_code, users_are_anonymous, activity_level;

Geoeditors Edits Monthly

The geoeditors_edits_monthly table is defined as:

column data type description
wiki_db string The wiki database the editors worked in
country_code string The 2-letter ISO country code this group of edits geolocated to, including Unknown (--)
edits_are_anonymous boolean Whether or not this group of edits were made anonymously
edit_count bigint Number of edits
namespace_zero_edit_count bigint Number of namespace zero edits
month string [partition] The month in YYYY-MM format

This is queried in similar fashion as geoeditor_edits. It is used primarily to report statistics to the Global Innovation Index (GII).

Unique Editors By Country Monthly

The wmf.unique_editors_by_country_monthly is similar to wmf.geoeditors_monthly, however, the data is not aggregated by wiki, thus allowing per country editor counts regardless of any particular project. The table is defined as:

column data type description
country_code string The 2-letter ISO country code this group of editors geolocated to, including Unknown (--)
users_are_anonymous boolean Whether or not this group of editors edited anonymously
activity_level string How many edits this group of editors performed

possible values: "1 to 4", "5 to 99", "100 or more"

distinct_editors bigint Number of editors meeting this activity level
namespace_zero_distinct_editors bigint Number of editors meeting this activity level, with only namespace zero edits
month string [partition] The month in YYYY-MM format

It is used primarily to report metrics and dashboards by the Global Data and Insights Team.

Archived report: Geoeditors edits yearly

A yearly report is generated presenting how many edits have been made by country and by wiki over the year. The report is split in two files, one contains edits made on all wikipedia projects, the other contains edits made on wikidata project.

Public bucketed version

A monthly report similar to the Geoeditors monthly, but made for public use.

Access

The easiest way to get to this data is by using Superset, a UI that you can use to access most of our datasources. The example geoeditors dashboard is at https://superset.wikimedia.org/superset/dashboard/9. As an example, here's a map generated from the data:

Screenshot of superset map generated with geoeditors data

Raw data

To compile the geoeditors_monthly dataset, we import MediaWiki's cu_changes table and filter out all the entries for administrative actions. This result is available as the wmf_raw.mediawiki_private_cu_changes table. We then aggregate at the day and user level, labeling anonymous users and generate the wmf.editors_daily table.

Since this raw data identifies the location of individual editors, we keep it for only 90 days, in accordance with our data retention guidelines. Data older than 90 days is continuously purged from the source cu_changes table, but since we regenerate the Data Lake's editing data every month, we instead keep data in mediawiki_private_cu_changes and editors_daily for the two latest calendar months (the month of the latest mediawiki_history snapshot and the previous). Older data may be temporarily available before it is purged, but you should not rely on this.

Editors daily

This can be found in wmf.editors_daily

column data type description
wiki_db string The wiki database of origin
country_code string The 2-letter ISO country code this group of edits geolocated to, including Unknown (--)
user_fingerprint_or_name[1] string For registered users, it is the global username, unique across all wiki dbs. For anonymous users, it is a hash of IP + user agent.
user_is_anonymous boolean Whether or not this user edited this group of edits anonymously
date string The YYYY-MM-DD date for this group of edits. Note that date is a reserved word in SQL, so you will need to quote it using backticks in Hive or Spark (`date`) or double quotes in Presto ("date").
edit_count bigint The total count of edits for this grouping
namespace_zero_edit_count bigint The total count of edits to namespace zero for this grouping
user_is_bot_by array<string> Whether this user is identified as a bot, values can be: group, name, both or empty
action_type int The action type for this group of actions - see https://www.mediawiki.org/wiki/Manual:Recentchanges_table#rc_type
month string [partition] The month in YYYY-MM format, all edits from that month from the cu_changes table are aggregated in this partition

wmf_raw.mediawiki_private_cu_changes

cuc_id          bigint   'Primary key, artificial',
cuc_namespace   bigint   'When pages are renamed, their RC entries do _not_ change.',
cuc_title       string   'When pages are renamed, their RC entries do _not_ change.',
cuc_user        bigint   'user_id, rev_user, the id of the user performing the action',
cuc_user_text   string   'user_text, rev_user_text, the name of the user at the time',
cuc_actiontext  string   'Unknown, undocumented',
cuc_comment     string   'The revision comment',
cuc_minor       boolean  'Whether this was a minor edit',
cuc_page_id     bigint   'The id of the page being edited',
cuc_this_oldid  bigint   'rev_id of the revision represented by this change',
cuc_last_oldid  bigint   'rev_id of the previous revision on this page',
cuc_type        int      'see https://www.mediawiki.org/wiki/Manual:Recentchanges_table#rc_type',
cuc_timestamp   string   'mediawiki-formatted timestamp of this event',
cuc_ip          string   'clear-text IP address of the user responsible for this event',
cuc_agent       string   'clear-text user agent of the user responsible for this event'

Changes and known problems since 2019-09

See also m:Research:Page view#Change log for changes to the page view definition itself, as well as Analytics/Data_Lake/Traffic/Pageview_hourly#Changes and known problems since 2015-06-16
Date from Date until Task record_version Details
2018-01 2019-08 T237072 n/a The count of namespace_zero_distinct_editors was calculated incorrectly for this period of time, for the "1 to 4" activity level. We replaced the invalid values with null values. All other activity levels and measures are fine for this period of time.

Data loss 2017

  • 2017: November 13th to November 20th, data not computed while the old database was decommissioned. Data will be lost if it's not re-computed by December 13th. As of this writing, we do not have capacity to save this data.
  • 2017: October 25th to November 4th, data not computed while changing database hosts. Data will be lost if it's not re-computed by November 24th (as of this writing, a few days have already been lost)

Notes

  1. 1.0 1.1 Until August 2022, user_fingerprint_or_id, that is locally unique to a wiki_db was used.