A significant tweeting practice we note in the #notracist dataset is the usage of more than one hashtag in a tweet - a phenomena we call Multi-Hashtagging.1) For the #notracist dataset, aside from the original #notracist term, there are a further 7717 hashtags, which are used in a variety of ways, for example:2)
We used Chorus to plot a 'cluster map' of key multi-hashtags (fig. 1). The Chorus software suite includes a Culture Explorer feature, which enables the co-occurrence of any terms within a Tweet to be identified.
In this map, each node (i.e. point in the map) is a hashtag, and the position and connectivity of nodes is determined by how frequently those hashtags are used together (co-occurrence), such that 'similar' hashtags cluster together (i.e. ones which are used in comparable ways to express similar sentiments). The model reveals how hashtags relate to each other as ‘semantic’ entities. The two radials overlaid on the image facilitate the process of reading the visualisation.
There is a tight central cluster of hashtags (including #funny and #lol) which are closely related to each other and demarcated within the inner (red) radial.
There are also a number of significantly populated (i.e. larger in area) nodes that feature on the outer branches extending from this central cluster (including #truth, #iswear, #fact and #justsayin/g, often appearing on the end of branches. These are demarcated in the outer (blue) radial.
The difference between the two radials is significant in as much they illustrate different tweeting practices, and we find that the inner radial consists largely of 'Humour'-type hashtags which are intended to mark tweet content as containing jokes or other comedic material; whilst the outer radial consists mostly of 'Truth'-type hashtags which tweeters use to clarify or qualify their statements by referring to them as observations and facts.
A further step in the analysis filtered the data into two sub-datasets representing 'Humour' and 'Truth' hashtags. To do this, we used the cluster map itself as a way of indicating which hashtags were associated with either category - broadly speaking, hashtags which featured in the inner radial were judged to be part of the 'Humour' category, and hashtags which featured in the outer radial towards the ends of topical branches were judged to be part of the 'Truth' category.3)
Having split our dataset in to the Humour and Truth categories, it was noticeable that both sub-datasets shared a basic lexicon – terms (words) such as 'black, 'white', 'people', 'like' and 'just' feature prominently in both 'Humour' and 'Truth' sub-datasets. It seems improbable that there is a linguistic or semantic means of consistently distinguishing between either category when attempting to interpret the meaning of a tweet. What is significant is that the words used in both 'Humour' and 'Truth' categories are broadly similar, and it is the tweeting practices associated with each category that accounts for their distinction. To demonstrate this, see the two images below which represent cluster maps plotting individual words contained within the 'Humour' (fig.2 ) and 'Truth' sub-datasets (fig. 3).
These cluster maps are visually different from each other - the 'Humour' map is a messy set of interconnected branching terms, whereas the 'Truth' map shows a high degree of term density around the far edges of the map (i.e. as far away from each other as the cluster map model will allow). The explanation for the different visualisations lies in the tweeting practices associated with each category. Typically, we find:
'Humour' hashtags are used to promote, propagate and share tweet content or internet objects (via URLs), i.e. a Vine video or Instagram pictures, and use many multi-hashtags to maximise the findability of the object or tweet content. E.g.:
In the 'Humour' category, multi-hashtags are frequently used together - this accounts for their tight central clustering visible in our original hashtag cluster map (fig. 1). These hashtags are drawn from a relatively narrow pool of terms which are repeatedly used. 'Humour' as a type of racialized talk relies on an implicitly-agreed-upon set of general classificatory hashtags which users recognise and drawn on in order to situate their tweets as embodying racialised humour (rather than actual racist intent). The 'Humour' hashtags themselves do not explain the meaning of the tweet, because the hashtags themselves - as a kind of self-referential meta-data4) - are the tweet. Humour-type tweeting appear to proliferate a repetitious cacophony of racialized meanings and affects.
'Truth' hashtags are, in contrast to their 'Humour' counterparts, used more sparingly and are drawn from a much wider pool of hashtag terms. Where the usage of 'Humour' multi-hashtags conforms to the well-understood function of hashtags as rendering something searchable, 'Truth' multi-hashtags are deployed differently as a means of clarifying or qualifying the semantic content of tweets. For example:
Here, the usage of multi-hashtags indicates how the tweeter intends for their tweet to be interpreted - their 'stance' 5) - as not representing any racist intent (i.e. #notracist) and justifying this disaffiliation with racism because the tweeter is stating what they argue is a defensible or observable everyday truth (e.g. #justsayin). This gives the 'Truth' cluster map (fig. 3) its distinctive outer density pattern - the wide variety of largely non-associated terms appear as distantly related from other, due to their lack of co-occurrence together. ‘Truth’ as a type of racialized talk relies on a diverse array of largely single-use hashtags which users draw on not to participate in wider debate or discussion, but to make explicit their semantic intentions (however misplaced or ignorant).
A user adding a truth-type hashtag alongside #notracist seeks to stabilize the ambiguity of racial meaning. Yet the creation of many singular truth-type hashtags makes such a practice a fraught activity. The proliferation of different truth-type hashtags is symptomatic of the dissonant registers of how racism-denial is mobilised in everyday online discourse.
We find that racialized hashtagging on Twitter as a mode of denial is not principally located in the variation of language within tweets, but in the classification of tweets by way of hashtagging. And hashtagging needs to be grasped as a techno-cultural practice of Twitter, rather than simply determined by individual users.
The ‘master’-hashtag #notracist, racially charges other hashtags by activating differential modalities of racist expression. In this respect, race as a digital assemblage is not simply inscribed in Twitter messages, nor can it readily de-code their meanings. Rather, racialized expression is an immanent process, a reactive force that emerges in a Twitter assemblage through hashtagging practices.