Research Paper Using Open Measures Data Reveals a New Use Case
NYU researchers studied how Kiwi Farms harassment attacks were sustained through the site's anti-abuse systems. This reveals a new use case for Open Measures tools.
TLDR
NYU researchers used data from Open Measures to investigate sustained Kiwi Farms harassment attacks. More than 25% of targeted threads received replies for longer than 12 months.
Most sustained Kiwi Farms harassment attacks followed a “continuous narrative escalation” model. In this model, new narrative elements added to threads encourage networked users to sustain them.
When the subjects of Kiwi Farms harassment sent take-down requests, the site posted them publicly. In turn, this often led to increased harassment.
When targets ask third parties, like infrastructure providers, to remove content, Kiwi Farms publicly posts these requests. This policy also applies to sensitive information, such as unedited abuse reports, putting harassment victims at risk.
The paper shows a new way to use Open Measures tools: studying anti-abuse systems through user behavior.
Background on Kiwi Farms Harassment
Kiwi Farms is an online forum dedicated to harassment and is one of the datasets collected by Open Measures. Journalists at Le Monde1 and NY Mag2 have previously reported on Kiwi Farms harassment campaigns and their impacts.
In August 2022, concerned streamers mounted the #DropKiwiFarms campaign3 to urge the site’s service provider, Cloudflare, to revoke its services. On Sept. 3, 2022, Cloudflare capitulated and blocked the site, drawing mainstream media attention.4 While Kiwi Farms was temporarily down while seeking alternative service providers, it has since come back online.
Because of its culture and content policies, Kiwi Farms harassment threads tend to last longer than those on image boards like 4chan. Open Measures collaborated with researchers at NYU to understand why and how Kiwi Farms harassment attacks were so sustained.
The resulting research paper was published in the April 2024 issue of Proceedings of the ACM on Human-Computer Interaction. This blog provides an overview of the methodology and key discoveries outlined in the paper.
Analysis of Kiwi Farms Harassment
To understand Kiwi Farms’ harassment attacks, researchers focused on the site's targeted subforums (dedicated areas discussing specific targets). To refine the dataset, they used additional filters to detect names in thread titles and messages. The search resulted in approximately 12,000 targeted threads.
Researchers discovered that 25% of threads in their dataset stayed active for over a year, and 30% got more than 100 responses. These statistics indicate that Kiwi Farms harassment threads are not short-lived or isolated. As a result of the site’s content policies and structure, new responses reactivate threads, generating more activity.
From their targeted dataset, researchers randomly selected 231 threads for more comprehensive, qualitative analysis. For range of comparison, they also selected a set of 100 threads from a separate subforum discussing recent take-down requests.
Findings
Narrative Elements in Kiwi Farms Harassment Threads
Researchers first analyzed user behavior to understand why certain Kiwi Farms harassment campaigns continued. They found that thread creators typically introduced a target in the first post to establish an initial attack narrative.
NYU researchers noticed that most posts contained or consisted of three elements: identity, content, and activity. They further defined “narrative elements” as any addition to a thread containing one of the above elements.
Identity refers to post content defining a target’s characteristics such as race, gender, religion, and ideology. In the data, these details were conveyed both as neutral background information and in other cases as hate speech. Below is one example of the identity element [sic, redacted]:
“[Target] identified as a "Turbofeminist" and white knighted [Gamergate key character]
and others during the beginnings of the gamergate shitstorm.”
In this case, the identity component of the post was the target’s name, ideological beliefs and background context.
Content refers to external links or embedded audio-visual material within a post (not in the sense as “content” meaning "words in a post"). This narrative element was the most common in attack threads targeting individuals.
Forum participants stress the importance not only of collecting what NYU researchers term “content” but also archiving it. This highlights a community practice on Kiwi Farms: backing up thread content in case the original links or files go offline (making sustained harassment easier). For example [sic]:
“If possible, see if you can download + archive all of the videos in question. She seems like the kind of person who will eventually take them down.”
Further, the focus on archiving content enables Kiwi Farms harassment threads to continue, as users can return to threads much later with all original reference materials still available. Researchers found that in their subset of 231 threads, there were 101 instances of “thread revivals”—attack threads that picked up after temporarily stopping for a month or longer. They found that about 62% of “revival messages” to reignite these threads included a narrative element of identity, content, or activity (in various combinations).
Activity, the third narrative element, refers specifically to any references made to a target’s online or offline statements, speech, or actions, specifically those used by attackers to expand their narratives against a target. Typically, user networks in Kiwi Farms harassment threads tended to concentrate on a target’s online speech and statements, since those were most readily available. For example:
“[Online handle of target] was [X-kind] of blogger and collector who liked to stir up drama on [Y] blogs and forums.”
Take-Down Requests Escalating Kiwi Farms Harassment
Take-down requests are formal demands made to remove specific content from the internet. Researchers found that due to the severity of their harassment, many of the targets of Kiwi Farms harassment threads understandably attempted to seek content removal.
This was sometimes done by reaching out to Kiwi Farms directly, though in other cases, targets made requests through intermediaries (infrastructure providers, third-party legal representatives). When requests went to third parties, they were forwarded back to the site, where portions were then published in a subforum dedicated to take-down requests. Researchers found that through this system, take-down requests functioned as reminders of old threads, thus reactivating user activity.
The following post is a clear example that a target’s take-down request led to the revival of the original thread, as explained by a user involved in the target’s harassment [sic]:
“I thought this thread was dead boring until I started to follow links and stuff. Thanks, crazy hippie lady, for threatening us with a lawyer, making me take a second look!”
Examples like the above show that the publication of take-down requests can trigger network responses, resulting in Kiwi Farms harassment becoming ongoing and sustained for certain targets.
Kiwi Farms Harassment through Continuous Narrative Escalation
Researchers found that the initiation and continuation of targeted Kiwi Farms harassment threads (and subsequent attacks) were prompted by the introduction of new narrative elements related to identity, content, and activity, and/or by take-down requests received by the site. These elements contribute to an ongoing cycle of online harassment that researchers defined as "Continuous Narrative Escalation."
First, a thread creator creates a post including one or more narrative elements (identity, content, and/or activity) to direct attention to a target. The post then generates user discussion, resulting in a “network verdict” about whether or not to proceed and how (i.e., is the target worthy of further harassment, what strategies will be used, etc.). When these Kiwi Farms harassment campaigns proceed, the network responds by continually expanding the narrative by adding new narrative elements through a process termed continuous narrative escalation.
Finally, these threads can fizzle for a time if an attack narrative can’t be productively expanded with new elements, but can always easily be reignited if new narrative elements can be added later (particularly because all original content is archived). Similarly, harassment may resume if a target requests that inflammatory content be taken down (since doing so redirects user attention to the forgotten thread).
Conclusion
The NYU researchers’ findings are helpful to other researchers and OSINT practitioners in several ways. First, their research provides a helpful analysis of the culture of Kiwi Farms harassment, both in terms of user behavior and site structure. Secondly, it offers the model of “continuous narrative escalation” as a helpful lens through which to understand the structure of other online harassment campaigns and how they might be interrupted.
Finally, the continuous narrative escalation model posited by the paper also opens up new areas of potential research and analysis: how third-party reporting systems may ignite abusive behavior (whether intentionally or inadvertently). Here, the researchers have found a new application of and use case for Open Measures crawling tools.
If you have any research topic suggestions for Open Measures’ researchers, reach out to us below:
Le Monde. (2022, September 5). Decoding kiwi farms, the blocked internet bully that continues to be a nuisance. Le Monde.fr. https://www.lemonde.fr/en/pixels/article/2022/09/05/decoding-kiwi-farms-the-online-bully-that-continues-to-thrive-despite-being-blocked_5995916_13.html
Pless, M. (2016, July 19). Kiwi Farms, the web’s biggest community of stalkers. Intelligencer. https://nymag.com/intelligencer/2016/07/kiwi-farms-the-webs-biggest-community-of-stalkers.html
Gonzalez, O. (2022, September 19). Kiwi Farms: What to know about the hate-filled message forum dropped by Cloudflare. CNET. https://www.cnet.com/culture/internet/kiwi-farms-what-to-know-about-the-hate-filled-message-forum-dropped-by-cloudflare/
Tiku, N. (2023, September 3). Inside the fight to keep Kiwi Farms, an anti-trans website, offline - The Washington Post. Washington Post. https://www.washingtonpost.com/technology/2023/09/03/kiwifarms-website-offline/