Authors:
(1) Anh V. Vu, University of Cambridge, Cambridge Cybercrime Center ([email protected]));
(2) Alice Hutchings, University of Cambridge, Cambridge Cybercrime Center ([email protected]));
(3) Ross Anderson, University of Cambridge and University of Edinburgh ([email protected]).
Ties
Summary and 1 Introduction
2. Travel and impact
2.1. Related work
2.2. Kiwi farmers
3. Methods, data and ethical sets, and 3.1. Forum and image table discussions
3.2. Telegrams cats and 3.3. Traffical and research on web traffic and analyzes
3.4. Tweets made by the online community and 3.5. Data license
3.6. Ethical considerations
4. The impact on forum activity and traffic, and 4.1. The impact of major disturbances
4.2. Displacement of the platform
4.3. Traffic fragmentation
5. The impacts on the stakeholders concerned and 5.1. The community that launched the campaign
5.2. Industry responses
5.3. Forum operators
5.4. Forum members
6. Tensions, challenges and implications and 6.1. The effectiveness of the disturbance
6.2. Censorship against freedom of expression
6.3. The role of industry in content moderation
6.4. Political implications
6.5. Limitations and future work
7. Conclusion, thanks and references
Annex A.
3. Methods, data and ethical sets
Our main method is based on data, with results supported by quantitative evidence derived from several longitudinal data sources, which we regularly collect. When quantitative measures require enrichment – as when analyzing the relevant public declarations of technological companies directly involved in the disturbance and ads made by forum operators – we use qualitative content analysis.
3.1. Forum and image table discussions
In addition to common current social networks like Facebook and Twitter, independent platforms such as Xenforo[4] and infinity[5] have gained popularity as tools to build communities online. Although they are less visible and require more maintenance, these can offer greater resistance against external intervention because operators have total control over the content and databases, allowing easy backup and redeployment in the event of a disturbance. These platforms generally share a hierarchical data structure ranging from babbles to threads linked to specific subjects, each containing several publications. While facilitating freedom of expression, they nourish and also diffuse more and more hatred and abusive discourse. We scratch the two most active forums associated with online harassment for years due to their increasingly toxic content, as part of the EXTREMEBB data set [62]: KIWI FARMS and LOLCOW FARM.
Our collection includes not only publications but also associated metadata such as publication time, user profiles, reactions and levels of toxicity, identity attack and threat measured by the Google perspective API in January 2023.[6] The API Perspective also offers other measures such as insult and blasphemy [63]But we exclude them because of the lack of relevance for the purpose of this document. This API uses crowdsourced annotations for model training and considerably surpasses alternatives [64]. We strive to ensure the exhaustiveness of the data by designing our scrapers to visit all the subfums, wires and messages while keeping a trace of each progress of the ramp to gradually resume in the event of interruption. A summary of forum discussion data is presented in Table 1.
Kiwi Farms is built on Xenforo, but operators have maintained the forum with their own efforts since the end of 2021, when Xenforo officially revoked their license. Our data covers all the history of the forum in early January 2013 at the end of 2022 with publications of 10.1 m in 48.3K Threads manufactured by 59.2K active users, offering a complete landscape through its evolution over time. While some extremist forums have experienced fluctuating activity and rapid decreases in recent years [62]Kiwi Farms showed stable growth until it was considerably disturbed in 2022 (see Figure 1). Our data precisely captures the suspensions reported, including those in 2017 and 2022.
The main rival of Kiwi Farms is Lolcow Farm, an image table built on infinity [65],, [66]. While discussions on Kiwi farms are largely based on the text, Lolcow Farm is focused on descriptive images. While Kiwi Farms users adopt pseudonyms, users of the Lolcow farm remain mainly hidden under the unified “anonymous” handle. We gathered a complete snapshot from Lolcow Farm from its creation in June 2014 at the end of 2022, encompassing 4.6 million messages made in wires of 10.0k. Lolcow Farm has much fewer threads, but everyone generally contains a lot of messages. This collection bears the total number of positions for the two forums at 14.7 m (and still growing). We exclude Lolcow, a smaller competitor from Kiwi Farms (also based on Xenforo), because he disappeared in the middle of 2022 and had less than 30,000 positions in total. As Lolcow Farm is now the biggest competitor, the analysis allows us to estimate the movement of the platform when Kiwi Farms was broken.
3.2. Telegrams cats
During periods of inaccessibility, the activity level increased in telegram groups associated with Kiwi farms. There are two channels: one is mainly used by forum operators to broadcast announcements and updates, especially where and when the forum could be accessible; And one is adopted by forum users mainly for normal discussions. The two channels allow public access, allowing people to join and visualize historical messages. We used the Telethon[7] To collect an instantaneous of these channels throughout their lifespan until the end of 2022, including 525K of messages, 298K of responses and associated metadata such as the number of views and 356K emoji reactions made by 2,502 active users. The data is probably complete because our scraper works in time almost real, and messages with metadata is fully captured thanks to the use of official telegrams API. Since forum operators are very encouraged to keep users quickly informed, their ads provide an incident and reliable response calendar.
3.3. Traffical and research on web traffic and analyzes
We found from ads in the Telegram group that Kiwi farms could be accessible via six main areas: the main thing is kiwifarms.net and four alternatives are kiwifarms.ru, kiwifarms.top, kiwifarms.is and kiwifarms.st, while a decentralized web version of Plerome is in kiwifarms.cc.[8] To study how users have sailed in these areas when the forum has experienced a disturbance, we have analyzed traffic analysis towards the six areas provided by similar – the main market platform providing information and intelligence in web traffic and performance.[9] Their reports include anonymous statistics from several inputs, including their own analytical services, sharing data from ISPs and other measurement companies, data has crawled from billions of websites and peripheral traffic data (website and application) such as plugins, complementary modules and pixel monitoring. Their algorithm then extrapolates the substantial aggregated data from the entire Internet space. Their estimate may not be completely precise, but reliably reflects trends at global levels and countries. To test this reliability, we have deployed our own infrastructure to collect more than 19 million traffic on the ground over six months, grouped them in 30 -minute sessions, then compared to similar visits. We note that, while underestimating the quantity of traffic because of the way in which the rehearsal pages are counted, the similarity is able to capture trends with a strong positive linear relationship (pearson r = 0.83 correlation coefficient). Our analysis in the following section also suggests a strong correlation between traffic data and forum activity
As similar does not offer an academic license, we use a free trial account[10] To access the longitudinal data of web traffic and the commitment that go back in the last three months. This includes information on total visits, unique visitors, the duration of the visit, the pages by visit, the rebound rate and the view pages. It also provides figures on research activity, marketing data such as visiting sources (for example, direct, research, email, social, reference, advertisements) and a non-temporal overview of geography and demographic data from the public. These data, covering office and mobile traffic, offer valuable perspectives. They extend from July to December 2022, two months before and four months after the disruption; This period is sufficient because there has been no significant intervention of the industry against the forum in the past (as shown in Figure 1), and the disturbance campaign has mainly ended after a few months (see §4). In addition, we have also collected research trends by countries and territories over time from Google Trends, covering the entire life of the forum. These two data sets are likely to be complete because they were collected directly from similar and Google.
[4] The Xenforo platform: [5] The Infinity Imageboard: https://github.com/ctrlctrlv/infinity/ [6] Google API Perspective:
