With Historical PowerTrack, Tweets are written to the archive as they are posted. However, when the archive was built in 2012 it included Tweet JSON that had been normalized and backfilled to some extent. For example, the “entities” structure that contains hashtags, mentions, and symbols was built out for periods before those entities official existed (see here for more background). Furthermore, some metadata, such as whether an account is verified, was backfilled. Accordingly, if you query for early Tweets from 2007, you’ll find user profiles that are marked as verified, even though account verification did not begin until 2009.
As mentioned in our documentation, there are important metadata details about the Historical PowerTrack archive:
- URLs: The
url_contains:operator will still function prior to 3/26/2012, but will only match against URLs as they are entered by a user into a Tweet and not the fully resolved URL (i.e. if a bit.ly URL is entered in the Tweet it can only match against the bit.ly and not the URL that has been shortened by bit.ly)
- Geo: Native geo data prior to 9/1/2011 is not available in Historical Powertrack. As a result, all operators reliant on this geo data will not be supported for jobs with a timeframe prior to this date.
- User Profile Data: All data prior to 1/1/2011 contains user profile information as it appeared in that user’s profile in September 2011. (e.g @jack’s very first Tweet in March 2006 contains his bio data from September 2011 that references his position as CEO at Square, which was not in existence at the time of the Tweet). User profile display name, 'home' location, and bio description information reflects the values of the referenced user at the time of the Historical PowerTrack job processing.
- Followers and Friends Counts: All data prior to 1/1/2011 contains followers and friends counts equal to zero. As a result, any rules based on non-zero counts for these metadata will not return any results for a timeframe prior to this date. Starting in late November 2019 user profile information will reflect the values of the referenced user at the time of the Historical PowerTrack job processing.
- Twitter language classifications begin appearing in November 2013.
- Profile Geo metadata begins appearing on June 4, 2013. User profile display name, 'home' location, and bio description information reflects the values of the referenced user at the time of the Historical PowerTrack job processing.
Below is a timeline of when Historical PowerTrack Operators begin matching. In some cases Operator matching begins well after a ‘communication convention’ became commonplace on Twitter. For example, @replies emerged as a user convention in 2006, but did not become a ‘first-class’ object or event with ‘supporting’ JSON until early 2007. Accordingly, matching on @replies in 2006 requires an examination of the Tweet body, rather than relying on the
in_reply_to_status_id PowerTrack Operators.
The details provided here were generated using Historical PowerTrack test jobs, and were informed by the Twitter timeline provided HERE. This timeline is not 100% complete or precise. If you identify another filtering/metadata “born on date” fundamental to your use-case, please let us know.
- January 3 -
- January 30 -
in_reply_to_status_id:@Replies become a first-class event after becoming a user-convention in October 2006.
- April 1 -
#Operator. Hashtags become a common ‘organizing’ tool in August.
- February 27 -
@Operator begin matching.
- February 27 -
- September -
emojicodes start appearing in Tweet metadata
- January 1 -
retweet_of_status_id:. Retweets became a convention as early as April 2007, but matching before this date depends on search for “RT @” or “Via @” patterns.
- January 1 - Followers and friends counts begin to be non-zero.
friends_count:start having non-zero values to match on.
- June 2 -
- September 1 -
- March 26 - Gnip introduces new data enrichments.
- Gnip Language -
gnip.langlanguage metadata. No longer filtered for.
lang:Operator now based solely on root level Twitter language classification.
- Expanded URLs - URL metadata from this date until launch of HPT 2.0 will contain
gnip.expanded_urlfully unwound URL. URL matching before this date will be based on URL as entered by the user. If the Tweet includes a shortened URL, e.g. bit.ly, there will be no expanded URL to match on.
- Gnip Language -
- November -
lang:Operator (matching on Twitter language classification).
- June 4 -
- August 20 -
- February 10 -
has:videosbegins matching on videos shared through Twitter’s user-interface (‘native’ videos).
- September 28 -
is:quotematching on Quoted Tweets.
- July 28 -
url_description:supported via Enchanced URLs enrichment.
- January 17 - ‘quote_count’ and ‘reply_count’ available in enriched native format. No associated Operators for these metadata.
- February 22 - Poll metadata become available in enriched native format. No associated Operators for these metadata.
Given all the above timeline information, it is clear that there are a lot of details to consider when generating historical Tweet datasets. There are two key things to consider:
- Some metadata have ‘born-on’ dates so filters can result in false negatives. Such searches include Operators reliant on metadata that did not exist for all or part of the search period. For example, if you are searching for Tweets with the
has:videosOperator, you will not have any matches for periods before February 10, 2015. That is because that Operator matches on native videos (attached to a Tweet using the Twitter user-interface). For a more complete data set of video-sharing Tweets, filters for before Febuary 10, 2015 would need to contain rule clauses that match on common URLs for video hosting.
- Some metadata, such as user profiles, have been backfilled with metadata from a time after the Tweet was posted.
As discussed HERE there are several attribute types that are commonly focused on when creating PowerTrack queries:
- Twitter Profiles
- Original or shared Tweets
- Tweet language classification
- Geo-referencing Tweets
- Shared links media
Some of these have product-specific behavior while others have identical behavior. See below for more details.
With Historical PowerTrack, the profile is as it was at the time the Tweet was posted, except for data before 2011. For Tweets older than 2011, the profile metadata reflects the profile as it was in September 2011. User profile display name, 'home' location, and bio description information reflects the values of the referenced user at the time of the Historical PowerTrack job processing.
Original Tweets and Retweets
is:retweet Operator enables users to either include or exclude Retweets. Users of this Operator need to have two strategies for Retweet matching (or not matching) for data before August 2009. Before August 2009, the Tweet message itself needs to be checked, using exact phrase matching, for matches on the “@RT ” pattern. Also, if you are filtering on Retweets from between May-August 2009, the “Via @” pattern should be included. For periods after August 2009, the is:retweet Operator is available.
Tweet language classifications
For filtering on a Tweet’s language classification, Twitter’s historical products are quite different. With Historical PowerTrack, Twitter’s language classification metadata is available in the archive beginning in November 2012.
As discussed HERE there are three primary ways to geo-reference Tweets:
Geographical references in Tweet message. Matching on geographic references in the Tweet message, while often the most challenging method since it depends on local knowledge, is an option for the entire Tweet archive. Here is an example geo-referenced match from 2006 for the San Francisco area based on a ‘golden gate’ filter.
Tweets geo-tagged by the user. The available geo-tagging history is dependent on the Historical API you are using. With Historical PowerTrack, geo-referencing starts on September 1, 2011. When the Historical PowerTrack archive was built, all geo-tagging before this date was not included.
Account profile ‘home’ location set by user. As with Tweet geo, the methods to match and the time periods available depends on the Historical API you are using. With Historical PowerTrack, these Profile Geo metadata is available starting in June 2014. For Tweets posted before the Profile Geo metadata, the
bio_location:Operator is available which can be used to match on non-normalized user input. User profile display name, 'home' location, and bio description information reflects the values of the referenced user at the time of the Historical PowerTrack job processing.
Shared links and media
In March 2012, the expanded URL enrichment was introduced. Before this time, the Tweet payloads included only the URL as provided by the user. So, if the user included a shortened URL it can be challenging to match on (expanded) URLs of interest. With Historical PowerTrack these expanded URL metadata are available starting in March 2012.
In July 2016, the enhanced URL enrichment was introduced. This enhanced version provides a web site’s HTML title and description in the Tweet payload, along with Operators for matching on those. With Historical PowerTrack, these metadata become available in July 2016.
In September 2016 Twitter introduced ‘native attachments’ where a trailing shared link is not counted against the 140 Tweet character limit. Both URL enrichments still apply to these shared links.
Here are when related PowerTrack Operators begin matching:
- 2006 November 1 -
- 2011 June 2 -
has:mediafor native photos.
- 2012 March 26 -
url:with the Expanded URLs enrichment
- 2015 February 10 -
has:videosfor native videos.
- 2016 July 28 -
url_description:with the Enhanced URLs enrichment