In 2015, Yvette Cooper announced the Reclaim the Internet campaign. This was in response to the abuse, often misogynistic in tone, faced by high-profile women online. However, the ubiquity of social media in society today means that this abuse, or trolling, is not exclusive to public figures. The campaign therefore aims to challenge this widespread trolling, promote online safety and provide support to victims.
Previous research by Demos in 2014 sought to measure misogyny online. It found that, over a 26 day period, around 131,000 tweets sent by UK-based Twitter accounts contained the words ‘slut’ or ‘whore’ – that’s over 5,000 tweets per day. However, it is important to note that this is not an issue specific to Twitter, as misogyny is present across all social media platforms just as it is throughout wider society. Twitter simply facilitates research like this by providing an easy-to-access, substantial dataset in a relatively short space of time.
While our previous research showed that misogynistic language was common online, it was more difficult to infer how these words were actually being used. New research by Demos’ Centre for the Analysis of Social Media has therefore used a Natural Language Processing Algorithm, whereby a computer can be taught to recognise meaning in language like we do. This means that the different uses of ‘slut’ and ‘whore’ can be classified, allowing us to further illustrate the nature and extent of misogyny online.
To do this, we collected just under 1.5m tweets from around the world over a period of 23 days. Of these, over half (54%) were advertising pornography, a startling figure considering Twitter’s minimum age requirement, like many social media sites, stands at 13. This may conflict with the UK government’s proposed Digital Economy Bill, announced in the Queen’s Speech, requiring websites with pornographic content to verify that their users are over 18. With pornography seemingly so common online, websites such as Twitter may need to take measures to censor pornographic material more effectively or alternatively increase their minimum age requirement to 18, both of which arguably hinder the openness and inclusivity of social media.
The remaining 650,000 non-pornographic tweets were sent by 450,000 individuals, representing just over 28,000 tweets per day. Two types of language was classified: ‘aggressive’ and ‘self-identification’. 33% of the tweets analysed were considered aggressive, whereby the tweets contained further obscenities in addition to ‘slut’ or ‘whore’, included commands (e.g. ‘shut up’), or used ‘you’ to target a specific user. In contrast, just 9% were considered self-identifying, whereby users appeared to reclaim these words to talk about themselves (e.g. “I’m a slut for beautiful sunsets”) or in a jovial manner when directed towards others (e.g. “happy birthday little slut I guess I love you”). The tweets that were not considered aggressive or self-identifying were classified as ‘Other’, usually focussing on slut walks, slut-shaming or discussing the use and connotations of these words.
In total, we found 213,000 tweets containing aggressive uses of the words ‘slut’ or ‘whore’. This represents over 9,000 aggressively misogynistic tweets sent per day worldwide during this period, with 80,000 Twitter users targeted by this trolling. Interestingly, this study reflects the findings of our 2014 report, in which women were as comfortable using misogynistic language as men; the 2016 findings show that 50% of the total aggressive tweets were sent by women, while 40% were sent by men, and 10% were sent by organisations or users whose genders could not be classified. These figures suggest that misogyny is being internalised and reiterated by women themselves. This use of language is not, therefore, confined to one discrete online group but rather persists throughout society, making this issue more complex than it first appears.
10,500 aggressive tweets, targeted at 6,500 unique users, could be algorithmically located in the UK; this means that there are at least 450 tweets posted containing aggressively misogynistic language in the UK every day. It should be noted, however, that this is an extremely conservative estimate.
While this analysis is unable to highlight individual women’s experiences of online trolling, or definitively say how serious each case is, it does hint at the scale of aggressive misogyny online and shows why we need campaigns like Reclaim the Internet. While censoring this misogynistic content may seem like the easiest solution, emphasis should be placed upon education instead. Indeed, Ms Cooper advocates compulsory sex and relationship education in schools. This will promote respect between genders and equip individuals with the knowledge to confidently challenge misogyny rather than attempting to police it outright, whilst being supported by Reclaim the Internet. By allowing this freedom of speech to be exercised, misogyny can be openly confronted and discussed instead of being swept under the carpet of censorship.
Measuring something like Misogyny is an almost impossible task, and this study cannot claim to be exhaustive, for better or for worse. The methodology used was as follows.
Tweets were collected through Twitter’s stream API over 23 days: this brought in 1.4m tweets. These were then classified using Natural Language Processing Algorithms. NLP is inherently probabilistic, and with much of social media (and the words we use in general) context is everything. Nevertheless, the algorithms trained performed well. The algorithm used to classify pornographic tweets was 91% accurate (meaning that 91% of the time, it agreed with a human analyst that a tweet was advertising pornography). The classifier identifying aggressive tweets was 82% accurate, and the classifier identifying self-identifying tweets was 85% accurate.
It goes without saying that more research is required in this area, and many questions this leaves unanswered, but these results are strong preliminary findings.
The full write up of this piece of research is available here.