Monthly Report October 2024
This month, we are covering the impact of the 7th October Hamas-led attack on Israel on online discourse and the intersections of hate.
Data for October was based on 1,018,607 posts in 23 languages, across 6 social media platforms: Reddit, Twitter, 4chan, Gab, YouTube, and Facebook.
Content warning: Presented data may contain disturbing language related to online hate speech.
Average toxicity going down
The social media toxicity trend increased steadily up to October 8th, after which it gradually declined, reaching a low of 0.15. Overall, the toxicity levels for the period were lower compared to the previous month, when the trend remained stable at an average of approximately 0.24.
What is a toxicity level?
Toxicity levels represent the average toxicity score of all posts collected during a specific month.
Each message is given a toxicity score between 0 (safe) and 1 (extremely hateful), determined using lexicons of problematic words and phrases.
The spike in toxic messages during the first week of October may be linked to the one-year anniversary of the Hamas-led October 7th attack which remained a focal point of online discourse due to the ongoing conflict in the region.
Unexpected rise in toxicity levels against LGBTQ+ and Roma after October 7th
Following October 7th, toxicity toward Roma communities increased by an average of 0.18 points, while toxic messages against LGBTQ+ increased by 0.10 points, representing increases of 60% and 27.8%, respectively. Similarly, anti-migrant/refugees and sexist narratives also increased by 20%. Interestingly, we did not observe a significant rise in the average toxicity in messages related to Judaism or Islam.
However, it is important to interpret data concerning the Roma community cautiously, as smaller sample sizes in this baseline (number of messages: 7725) require thoughtful interpretation to account for potential variability in the data. This reflects our commitment to conducting nuanced and precise analysis when comparing trends across different communities.
Contrary to previous trends, the average toxicity levels targeting Jewish and Muslim communities did not increase around October 7th, despite the ongoing Middle Eastern conflict. However, we observed that terms associated with LGBTQ+ communities were used in a derogatory way when talking about Jewish or Muslim communities. Hence, discussions around polarising events may become more toxic through the misuse of language typically associated with the LGBTQ+ community. This highlights the interconnected nature of hate narratives.
Examples of high toxic posts in LGTBQ+ baseline.
German-language toxicity declines in October
All languages analysed, except Italian, experienced increased average toxicity following October 7th, while gradually decreasing towards the end of the month. Following heated conversations in September, German-language toxicity dropped in October from an average of 0.26 to 0.17. This shift indicates a calmer tone in October, suggesting a decrease in the intensity of online debates within German-speaking communities compared to the previous month. During September, several events may have contributed to heightened online tensions in Germany, including the far-right AfD party's electoral victory in Thuringia, temporary border restrictions, and a series of stabbing attacks in Solingen.
Conversations in the aftermath of October 7th were marked by increased circulation of hateful rhetoric surrounding the Middle East conflict, alongside deeply polarising debates about US political parties. These discussions may reflect both the emotional intensity of the anniversary and the broader divisive dynamics shaping public discourse.
Examples of English-speaking, highly toxic posts in the aftermath of October 7th
Lower toxicity across baselines compared to September
Our baselines faced a decreased average level of hate compared to the previous month. Roma people saw the largest drop, with hateful messaging decreased by 0.11 points compared to September.
Although average toxicity in antisemitism and islamophobia decreased slightly, the number of hateful posts discussing topics around Islam and Judaism (N=139,273 and N=161,202 respectively) increased compared to last month (N=103,428 and N=142,068).
This rise in post volume may be linked to the heightened focus on the ongoing conflict in the Middle East, which dominated global discourse and likely fueled increased engagement on these topics. The surge underscores how geopolitical events can amplify discussions steeped in hate, even as the overall toxicity of such messages remains relatively stable.
Fringe vs. mainstream: diverging toxicity trends following October 7th
Following October 7th, toxicity levels on fringe platforms spiked, maintaining a high toxicity average of around 0.4 until October 19th. In contrast, mainstream platforms maintained a stable average of around 0.2 throughout the month.
The elevated toxicity on fringe platforms following October 7th suggests that these online spaces not only amplify but also sustain heightened hate speech and polarising rhetoric in response to contentious events. This trend may reflect the nature of fringe platforms, where users are often less moderated and may seek spaces with less content moderation to express extreme or inflammatory views. Additionally, the prolonged period of high toxicity indicates that fringe platforms may serve as echo chambers, allowing hateful narratives to persist and escalate without significant counter arguments or interventions.
In contrast, mainstream platforms’ relatively stable toxicity levels point to differences in moderation policies and user demographics. Stricter content moderation, larger user bases with more diverse perspectives, and greater scrutiny from regulators and the public may all contribute to the ability of mainstream platforms to avoid the same spikes observed on fringe platforms. However, the disparity between fringe and mainstream toxicity trends raises questions about the spillover effects of fringe rhetoric and the extent to which these platforms influence broader online and offline discourse.