reportNovember 5 2018

Data Craft

The Manipulation of Social Media Metadata

Amelia Acker

Data Craft analyzes how bad actors manipulate metadata to create effective disinformation campaigns and provides tips for researchers and technology companies trying to spot this “data craft.”

Report Summary

“Data craft is about manipulating a system to assert power over it, and in doing so it can reinforce and reveal limits, or even blind spots in platforms.”

Data & Society Research Affiliate Amelia Acker’s Data Craft analyzes how bad actors manipulate metadata to create effective disinformation campaigns and provides tips for researchers and technology companies trying to spot this “data craft.”

“Data craft” is the term Acker coins for all those “practices that create, rely on, or even play with the proliferation of data on social media by engaging with new computational and algorithmic mechanisms of organization and classification.” Data Craft elucidates what metadata is, how users create metadata, and how metadata is used by platforms.

Bad actors typically manipulate “platform activity signals” like usernames, profile handles, bio fields, post dates, and follower counts. Acker unveils the “craftwork” of manipulating metadata with three case studies that focus on Instagram accounts of government representatives; official U.S. Government Twitter accounts; and Facebook ads purchased by the Russian-based Internet Research Agency.

Acker argues that understanding data craft is integral to identifying and combating disinformation campaigns, as data craft often allows disinformation campaigns to go undetected – especially by automated content moderation algorithms.

Intended for internet researchers, technology companies, and other stakeholders strategizing to combat disinformation, the report leaves readers with a concrete understanding of metadata’s role on platforms and avenues to identify vulnerabilities and hold platforms accountable.


Tips for Spotting Data Craft

On mimicking legitimacy
  • Search for accounts with the similar or same real name, user name, or account handle across other platforms
  • Compare profile pics and account descriptions across multiple platforms
  • Identify copies of posts and their time stamps
  • Reverse Google image search profile pics and account banners
  • Locate absences that have not been copied or backfilled (e.g., lack of comments, description)
  • Check if an account has been tagged by other verifiable platform users
  • Consider parody or organizational change as explanations before concluding that something is malicious manipulation
On detecting imposter accounts
  • Locate date of when account was started, user joined
  • Look at how many tweets/posts have been created since account start date
  • View attached media (pictures, videos, links) and look for duplicates
  • Consider the date of the last post or activity and if the account has been dormant
  • If it is listed as an official account, search for the personality/institutional homepage to cross-reference and confirm existence of an official account
  • Search the Internet Archive’s Wayback Machine for crawls of the account
  • Scan multiple crawls if possible and look to see if the account was dormant or has ever been deleted or suspended
  • Reverse image search the profile pics and banners to see if and where copies occur
  • Explore followers and commenters for their authenticity to assess if they appear to be real people or bots
  • Read to see if the comments are substantive and engaging with the content or are they simply reactions or emoji
Other strategies to examine metadata
  • Use social dashboards to see account creation time and date and average daily active posts, comparing the date of the account establishment and the posts per day
  • Examine promoted posts and ads policies of platforms, research
  • Consult the page administrator’s user account and page, comparing the rate of posting promoted content to free content
  • Examine how often content is shared (e.g., are memes or videos frequently re-shared but with different captions?)


Connected Track