The Relevance of Devices in Divergent Tweeting Practices

The first presenters on this second day at AoIR 2015 are Bernhard Rieder and Carolin Gerlitz, whose interest is in using data from Twitter's 'spritzer' firehose! which delivers a random 1% or all current tweets. How can this be used to identify individual types of activity in relation of the wider platform ecology? In particular, for the purposes of this paper, what light does it shed on the use of different devices for tweeting?

The project collected some 32 million tweets from the spritzer firehose over the course of one week, and key tools for tweeting were especially iPhone and Android devices. This may also be combined with the tweet contents themselves, to see which devices contribute especially strongly to specific hashtags, for example.

Another way to look at this is to examine the relative occurrence of different tweet elements per device. iPhone and Web clients frequently post @mentions, for example; Tweetdeck is stronger on hashtags and retweets; the automated tweeting tool Tweetadder is much stronger on links and hashtags, unsurprisingly; Instagram is very strong on links, for obvious reasons; while tweets from Facebook game Tribez always contain ions and hashtags. The iPhone is also very strong on teen hashtags (Miley, Selena, 5 Seconds of Summer), while Instagram hashtags are much more image-related, of course.

Bernhard now also shows the relations between tweeting languages, tweeting tools, and domains shared in links (or the other hashtags present) for specific hashtag datasets. This may show distinct differences between usage patterns of the users of different devices; it also points out the significant difference in usage patterns between different hashtags, or the way popular hashtags are hijack by spammers.

Different devices have different capacities, then, and point to different use practices. In news conversations, for example, professional Twitter clients are overrepresented; the iPhone, by contrast, is the preferred microphone of the American teenager, Carolin says. Custom posting clients (Instagram, games) are engaged mainly in 'activity loops': they send messages to Twitter but in their URLs point back to the original site. Some automation clients empower promotion, spam, and hashtag hijacking.

A closer look at news domains reveals some interesting patterns, too. Tweets linking to news URLs engage in a diverse range of hashtags, and automated clients especially connect to porn-related hashtags in order to give their bot accounts a more 'normal' appearance; tweets linking to YouTube often contained Arabic-language hashtags, although YouTube videos posted from iPhones largely participated in a teen idol hashtag. Tweets linking to DIY site Etsy, by contrast, we're especially posted from higher-end tools like Hootsuite and Roundteam, as Etsy users have trained each other to use such tools.

So, the devices used for tweeting intersect with Twitter practices; they are involved in different regimes of being on Twitter, and all tweets are thus not created equal. Tweets are not just prestructured and mediated by the Twitter platform, but also by the devices and tools used for sending, as they provide very divergent production environments.

This creates problems and opportunities for research, and the device origins of tweets should also be taken into account in developing overall tweeting metrics. 100,000 tweets from Tweetadder represent a very different dataset from 100,000 iPhone tweets, for example. This needs to be further explored and unpacked as we continue to develop our approaches and metrics for Twitter analytics.