Digital Library
Antisemitism on Twitter Before and After Elon Musk's Acquisition
Topic:
Antisemitism & Antizionism, Israel & Regional Politics
Principal Investigators:
Carl Miller, David Weir, Shaun Ring, Oliver Marsh, Chris Inskip, Nestor Prieto Chavana
Study Date:
2023
Source:
ISD,CASM Technology
Key Findings:
In October 2022, Twitter was acquired by Elon Musk at the head of a consortium of private investors. It precipitated one of the most dramatic shifts in social media’s short and tumultuous history of grappling with online harms as, within weeks, previously banned accounts were reinstated, policies were upended, and a significant proportion of Twitter’s staff were laid off, including reportedly many of those in the company responsible for online safety.
This paper looks to provide an initial snapshot of how these changes have impacted Twitter, through analyzing the scale of English-language antisemitism on the platform before and after the takeover. The study is neither exhaustive nor definitive, but it is a useful early window into the dynamics of one form of online hate and the responses to it.
Analysis shows the volume of antisemitic Tweets more than doubled after Musk’s acquisition. Between June and October 27th, the weekly average of plausibly antisemitic Tweets was 6,204. From October 27th until February 9, the average was 12,762, an increase of 105%.
A significant surge of new accounts posting plausibly antisemitic content was identified. 3,855 such accounts were created between Oct 27 and Nov 6, an increase of 223% compared to the 11 days (the equivalent timespan) leading up to Oct 27. Whilst Musk claimed that “hate Tweets will be max deboosted”, data showed only a very small decrease in the average levels of engagement or interaction with antisemitic Tweets before and after the takeover.
Analysts drew the following key themes of antisemitism from the Tweets collected:
-Conspiracist content, often referring to Jewish control of ‘elites’, media and politics.
-Antisemitic attacks on ‘Zionist states’ - including Israel, but also Ukraine - often tied to the above idea of Jewish ‘control’ of Western elites.
-Support for antisemitic comments made by Kanye West.
-Racialised antisemitism, including white nationalism, nativism and ethnosupremacism.
-Historical antisemitic tropes, including separating out historical ‘races’ of ‘fake’ and ‘real’ Jews and blaming Jews for the death of Jesus.
Methodology:
The researchers’ approach uses an innovative algorithmic architecture to classify Tweets which could be interpreted as ‘plausibly antisemitic’, where at least one reasonable interpretation of a message’s meaning fell within the International Holocaust Remembrance Alliance’s definition of antisemitism. There are inherent challenges in training language models on as nuanced a topic as antisemitism, but this architecture is evaluated to operate with an accuracy of 76%. This paper uses a pioneering approach to the automated detection of antisemitism by deploying an ‘ensemble’ of different classifiers within a single workflow.
Based on this criteria, between 1 June and 9 February 2023, a total of 325,739 plausibly antisemitic Tweets were identified as having been sent from 146,516 accounts.
The data used in this analysis was collected in two batches. The initial collection was performed on December 2, 2022 and included Tweets posted between June 1 to November 30. A second collection was performed on February 9 2023 that extended this range from November 30, 2022 to February 9, 2023. All analysis presented in this report refers to the full collection range, with the exception of the enforcement analysis which focuses on the initial date range only.
A combination of topic modelling and manual appraisal was used.