Online Library

The article explores Bitchute, a video-hosting platform associated with the Far/Alt Right, with the aim of understanding how it reconfigures political communication and the digital public sphere. Methodologically, the article employs the walkthrough method and non-participant observation to identify the main features and functionalities offered to users. These include a set of values that prioritise creators, an algorithmic organisation that keeps users engaged with a single creator channel rather than with the same topic across channels; and embedded buttons for tips and pledges for creators enabling them to directly monetise their content. The content posted on Bitchute tends to coalesce around politicised cultural issues. It is noteworthy that although Bitchute hosts some advertising, it does not use data for microtargeting and in general makes limited use of user data. We interpret these findings as suggesting that Bitchute constitutes a media infrastructure that encourages, incentivises and sustains microcelebrities of the Far/Alt Right, who act as ideology entrepreneurs. Bitchute can therefore be seen as an infrastructure for the multiplication/sustenance of ideological entrepreneurs/political influencers who vie for the attention and money of far-right publics. If we can speak of a structural transformation of the public sphere associated with Alt Tech, our discussion of Bitchute suggests that this takes the form of a political media infrastructure that enables the continued existence and consolidation of a new type of political actor, the ideology entrepreneur.

With the spread of social networks and their unfortunate use for hate speech, automatic detection of the latter has become a pressing problem. In this paper, we reproduce seven state-of-the-art hate speech detection models from prior work, and show that they perform well only when tested on the same type of data they were trained on. Based on these results, we argue that for successful hate speech detection, model architecture is less important than the type of data and labeling criteria. We further show that all proposed detection techniques are brittle against adversaries who can (automatically) insert typos, change word boundaries or add innocuous words to the original hate speech. A combination of these methods is also effective against Google Perspective – a cutting edge solution from industry. Our experiments demonstrate that adversarial training does not completely mitigate the attacks, and using character-level features makes the models systematically more attack-resistant than using word-level features.

With the spread of social networks and their unfortunate use for hate speech, automatic detection of the latter has become a pressing problem. In this paper, we reproduce seven state-of-the-art hate speech detection models from prior work, and show that they perform well only when tested on the same type of data they were trained on. Based on these results, we argue that for successful hate speech detection, model architecture is less important than the type of data and labeling criteria. We further show that all proposed detection techniques are brittle against adversaries who can (automatically) insert typos, change word boundaries or add innocuous words to the original hate speech. A combination of these methods is also effective against Google Perspective – a cuttingedge solution from industry. Our experiments demonstrate that adversarial training does not completely mitigate the attacks, and using character-level features makes the models systematically more
attack-resistant than using word-level features.

VOX-Pol Blog Post.

VOX-Pol Blog Post.